Sudden Stops and IMF Programs1 Barry Eichengreen, Poonam Gupta, and Ashoka Mody
December 2005 1. Introduction
The debate over the role of the International Monetary Fund in a world of capital
mobility shows no signs of dying down. There is a broad awareness that the single factor
that most distinguishes our current economic and financial environment from that of the
preceding period is high capital mobility. There is also a widely held view that while
exposure to international capital markets promises benefits, those benefits come packaged
with risks. But there is still no consensus on what the IMF should do about it. Should the
Fund reduce its role as emergency lender on the grounds that countries enjoy increasingly
free access to markets as an alternative source of finance? Or should it expand its role, since
relations with international financial markets expose countries to abrupt and potentially
costly reversals in the direction of capital flows? Is the existing range of IMF lending
facilities adequate for meeting these needs? Or is there a case for a new facility capable of
disbursing large amounts of financial assistance quickly, perhaps on the basis of
prequalification, to countries experiencing sudden shifts in the direction of capital flows?2
1 Prepared for the Interamerican Seminar on Macroeconomics, Rio de Janiero, 3-4 December 2005. The first author is with the University of California, Berkeley, the second and third with the IMF. The usual disclaimer, that none of the opinions expressed here are necessarily those of any institution with the authors are affiliated. We thank Marianne Schulze-Ghattas and Jeromin Zettelmeyer for comments and Sudarat Ananchotikul and Anna Unigovskaya for research assistance.
2 Two recent reviews of the debate which elaborate these points are Ostry and Zettelmeyer (2005) and Truman (2005).
- 2 -
More evidence may help to answer these questions. There exist substantial empirical
literatures on both shifts in the direction of capital flows (known by the moniker “sudden
stops”) and the effects of IMF programs. (See Tables 1 and 2.) Empirical analyses of
sudden stops focus on the impact on their incidence, magnitude and effects of economic
policies and characteristics like a country’s exchange rate regime, financial openness, and
dependence on international trade. Econometric analyses of IMF programs typically
examine the behavior of inflation, output, fiscal effort and, most relevant to the questions at
hand, the balance of payments, compared to a control group of country-year cases with no
program in place. But, to our knowledge, there exists no study focusing on the impact of
IMF programs on the incidence, severity and effects of sudden stops.
In this paper we take a first stab at developing such evidence. Our results suggest a
negative impact of IMF programs and credits on the likelihood of experiencing a sudden
stop. (We also find some evidence that the presence of an arrangement with the Fund
reduces the negative output effects of sudden stops.) This can be interpreted in terms of the
literature on global capital account shocks and the stabilizing effect of liquidity insurance.
Even countries with strong policies may experience a sudden curtailment of capital inflows
and a sudden shift to outflows if investors suspect that other investors, for whatever reason,
are primed to take their money out of the country. In such circumstances, emergency
financial assistance from the IMF can reassure individual investors of the country’s
continued ability to finance its international transactions and reduce their incentive to
scramble to liquidate their positions. Emergency lending by the IMF can ensure the
continued provision of private finance in much the same way that lender-of-last-resort
intervention by a central bank can limit the scope for bank runs.
- 3 -
We do not want to claim too much for these results. Complicated issues arise in the
course of analyzing the relationship between sudden stops and IMF programs. There is the
difficulty of measuring both IMF programs and sudden stops. There is the challenge of
identifying the impact of the former on the latter (specifically, of addressing the endogeneity
problem). At this point we have only provisional solutions for these and other problems. In
addition, our paper leaves open a variety of related issues, such as appropriate modalities for
lending to countries in this position, the feasibility of identifying qualifying countries in the
relevant time frame, and moral hazard. While these are not our topic in this paper, this does
not mean they are unimportant. Still, we believe that our results here are the first evidence in
the literature of the insurance properties of IMF programs.
The remainder of the paper is organized as follows. Section 2 reviews the debate
over IMF insurance against sudden stops. Section 3 then introduces our sample of sudden
stop episodes. Section 4 next describes our data on IMF programs. Section 5 sets up the
benchmark multivariate analysis, while Section 6 discusses strategies for identifying the
impact of IMF programs. Section 7 reports a series of robustness checks and extensions.
Section 8, in concluding, returns to the question of what implications we can draw for the
efficacy of IMF programs in reducing sudden stops.
2. Sudden Stops and Multilateral Insurance
The observation that sudden stops also cluster in time, as we shall see below, is taken
as suggesting that they have as much to do with the behavior of global financial markets as
well as country policies, which some commentators take to mean that their incidence would
be unlikely to increase as a result of the provision of multilateral insurance. Calvo (2005)
- 4 -
takes the fact that growth recovers quickly from sudden stops as further evidence that country
policies are not at the root of this phenomenon. He concludes that sudden stops and the
increasing frequency of financial crises in emerging markets reflect inefficiencies in
international financial markets and argues for emergency financial assistance to countries
suffering from sudden stops.3
A number of questions can be raised about this argument. A first one is why
countries cannot obtain insurance commercially by establishing credit lines on international
capital markets or issuing securities with embedded options that have the same insurance
properties. If the argument for insurance is strong, then the private sector should be prepared
to provide it for a fee. This objection seems especially compelling in light of the growth of
international financial markets and transactions in recent years. There has in fact been some
experimentation with this practice, by inter alia Argentina in the 1990s.4 But the Argentine
credit line was small, and execution was delayed until well into the crisis.
One possible explanation for why the practice is not more widespread is that capital
requirements and other regulations prevent potential suppliers from providing insurance on
the requisite scale. Commercial counterparties may also be worried about concentrated
country exposures and demand a prohibitively price for the provision of contingent credit
lines. Still, if the case for private insurance is strong, one presumes that financial markets 3 Calvo’s preferred variant of the mechanism would have the stabilization fund purchase the bonds of adversely-affected economies to prevent their spreads from rising (Calvo 2002).
4 The experiment in question involved a contingent repurchase contract between the Argentine central bank and a consortium of foreign banks, under which the central bank was allowed to withdraw funds in the event of a crisis via a three-month renewable credit line collateralized by dollar-denominated government bonds.
- 5 -
and institutions adept at diversifying and repackaging risks would find a way around these
obstacles.
Another possibility is adverse selection. If asymmetric information prevents potential
insurers from discriminating among borrowers in different risk categories, then only risky
countries will wish to contract for such lines. The higher are the fees and interest rates
charged, the greater the riskiness of willing purchasers, causing the private market to
collapse. The limitation of this argument is of course that potential insurers are far from
ignorant of variations in country risk. While information is asymmetric, in other words, it is
not all that asymmetric; lenders should still be able to charge different rates to borrowers
subject to different levels of country risk.
A related argument is that a public insurance agency may have more ability or
stronger incentives to gather information on the financial condition of its clients, in turn
enabling it to better tailor incentive-compatible contracts. The obvious objection here is that
private financial institutions with their own financial returns and performance at stake have at
least as strong an incentive to invest in these monitoring functions.
Yet another possibility is that commercial insurance providers have an incentive to
take a short position against the country when the probability rises that it may wish to draw
down its credit line. In turn, this will destroy the effectiveness of the insurance (Broda and
Levy-Yeyati 2003). Presumably this problem does not carry over to public insurance
providers.5
5 This problem would be ameliorated if the insurance liability was securitized and widely distributed, since diversification would then provide the insurers with the protection they need (obviating the need to hedge on a large scale). But in turn this begs the question of why
(continued)
- 6 -
A second question is whether bunching -- that multiple countries tend to experience
sudden stops simultaneously – limits the feasibility of multilateral insurance. If a substantial
subset of IMF members needs to draw on the resources of the Fund simultaneously because
they experience shocks simultaneously, then the financial feasibility of an insurance
arrangement can be questionable.
A third question is whether sudden stops are really not a function of country policies.
Empirical analyses of their incidence from Calvo, Izquierdo, and Mejia (2004) to Edwards
(2005) identify roles for both internal and external factors. To quote Calvo (2005),
“econometric studies do not reject the hypothesis that Sudden Stops are largely prompted by
external factors but, at the same time, strongly suggest that the probability of Sudden Stops
reflects domestic characteristics” In other words, even analysts emphasizing the importance
of global factors acknowledge that domestic characteristics shape the impact and response to
external shocks. Insofar as the relevant characteristics include policies under the control of
the domestic authorities, this means that the moral hazard problem must be addressed.6
emerging markets find it hard to place innovative securities containing put and call options that kick in, reducing debt service or even calling for reverse payments, when economic conditions deteriorate.
6 The extent and economic importance of moral hazard in this context is disputed; for a review of the evidence see Lane and Philips (2000). There is also the possibility that, in the presence of other distortions, adding insurance will lead to less risk taking rather than more. Thus, Cordella and Levy Yeyati (2004) show in a model of finite-lived governments where the probability of government survival declines with a rising incidence of financial crises, that insurance that reduces the risk of financial collapse may in fact encourage the authorities to invest more in policy reform (in the present context, to reduce the riskiness of their policies). Of course, there are actually two effects of insurance here: insurance that reduces the likelihood of a crisis reduces the pressure on the government to head it off, but that insurance also strengthens the incentive to pursue reforms that reduce the riskiness of the environment in the future. Predictably, the net effect is ambiguous. Cordell and Levy Yeyati
(continued)
- 7 -
Moral hazard does not render insurance infeasible, but it requires that a reasonable
insurance scheme be designed to limit its extent. An obvious way of doing so is through
surveillance and conditionality. A key question is whether such conditions are better applied
ex post or ex ante. Specifically, should the IMF announce in advance what countries are
eligible for a credit line in the event that they experience a sudden stop and specify the
amount of assistance that they can expect to receive? Or should it proceed on a case-by-case
basis and decide whether or not to provide additional credit once the sudden stop and the
severity of the output decline have been observed?
Those who argue for insurance against sudden stops generally favor ex-ante contracts.
First, the problems of illiquidity that arise when capital flows are interrupted can lead to
problems of insolvency unless funds are disbursed quickly. Determining eligibility with the
requisite speed requires that countries be deemed eligible for assistance ex ante. Second, if
the terms and amounts of external assistance are specified ex ante, then the government has a
stronger incentive to take steps to expedite the economy’s recovery from the sudden stop.
Cohen and Portes (2004) describe an insurance contract in which countries prequalify
for assistance if their debt ratios remain below a critical ceiling consistent with moderate
spreads, say 400 basis points above LIBOR. By assumption, any crisis that the country then
experiences is a crisis of liquidity, not a crisis of debt sustainability. Payments are triggered
when spreads on the debt rise above the threshold level. The IMF would then lend to the
country at the threshold spread. It would thus contain the effects of the sudden stop that show that insurance is more likely to encourage reforms that pay off in good times (since it reduces the risk of falling into a crisis in the first place) than reforms that reduce the risk of a crisis in the event that one occurs.
- 8 -
caused spreads to rise and prevent the liquidity crisis from degenerating into a solvency
crisis. Cordella and Levy Yeyati (2005) propose a country insurance facility that would
provide eligible countries with automatic access to a credit line at a predetermined interest
rate, where eligibility criteria would again focus on debt sustainability – not just the level of
the debt but also its maturity and currency composition. Dervis and Ozer (2005) similarly
propose a Stability and Growth Facility that would provide insurance against unforeseen
shocks and for which countries would prequalify on the basis of their policies.
Chami, Sharma and Shim (2004) provide a model laying out the analytics of this
approach. They assume that the insurer has two objectives: safeguarding its assets and
providing for the borrowing country’s welfare, which it can enhance by extending a loan. A
governmental counterparty decides in each of two periods how much unobserved effort to
exert in order to avoid and then recover from a financial crisis, which in turn affects how
much will be asked to repay to the insurance pool.
In this model it is preferable for the insurer to specify eligibility for assistance and the
terms of the credit line – how much assistance the insured will receive as a function of the
severity of the decline in output that occurs in the first period – before the fact rather than
deciding ex post whether or not to extend a loan after the country enters the crisis but before
the full output consequences, which only develop in the second period, are known. There is
moral hazard under either contract; knowing that it stands to receive official support, the
efforts of the national authorities to avert the crisis are correspondingly less. But the extent
of the moral hazard affecting the authorities’ efforts to recover from the crisis varies with the
form of the contract. Under the ex-post contract, the crisis country repays more if it makes a
greater effort to recover from the crisis, which works to depress adjustment effort. In
- 9 -
contrast, under the ex-ante contract, when terms are agreed prior the government’s decision
of how much adjustment effort to extend, repayment is independent of effort; hence
adjustment will be greater, and outcomes will be superior. In general, an insurer that both
has a fiduciary responsibility to safeguard its assets and that cares about the welfare of the
crisis country will prefer an ex ante contract in which the operator of the reserve pool
specifies who is eligible and the terms and amounts of the assistance that will be
forthcoming.
But there is a problem of time consistency with the ex ante contract. The insurer may
want to renege on its fixed commitment if it observes that the recession is unusually severe.
Since it values the welfare of the insured, it may then wish to offer more generous terms.
Hence ex ante contracts in which the amount and terms of the credit line are fixed will not be
credible. And since the insured knows this, moral hazard leading to less adjustment effort
will still be a problem, surveillance and conditionality or not.7
This suggests that the IMF may wish to deem certain countries eligible for a fixed
credit line, since this is better than a discretionary, case-by-case approach at preserving its
own financial solvency while at the same time supporting the welfare of its members, but that
the time consistency problem may undermine the feasibility of this approach and aggravate
moral hazard. And if moral hazard is serious, such an insurance arrangement may turn out to
be welfare reducing rather than welfare improving for the membership as a whole.
7 Theoretically, it is possible that this problem could be solved through repeated interaction between the insurer and the country, through which the former develops a reputation for acting consistently.
- 10 -
Two further questions can be raised about these ideas. First, doesn’t the
unsatisfactory experience of the IMF’s contingent credit line (CCL), for which no country
applied prior to its expiration in 2003, raise questions about the enthusiasm of countries for
ex ante insurance? Insofar as Executive Directors have a responsibility for the IMF’s own
solvency, members running risky policies that might prevent them from paying back credits
would have to be denied the privilege of borrowing. Such an outcome would send a negative
signal to the markets. This seems to have been what deterred governments from applying for
a CCL. A further problem was that eligibility could be rescinded at some future date,
sending a negative signal that might precipitate a crisis.
The Fund could meet its fiduciary responsibility by announcing unilaterally which
member countries were eligible for financial assistance. But it would then send a negative
signal about the financial condition of the other countries when it declared them ineligible for
insurance. In turn this might render countries reluctant to participate in the insurance
arrangement in the first place. And the exit problem would remain.8
Finally, such schemes tend to assume the ability of the agency operating the facility
to discriminate between solvent and insolvent countries, where the solvent countries are still
susceptible to liquidity crises and should thus be made automatic beneficiaries of the new
facility. Cohen and Portes (2004) assume the existence of a well-defined amount of debt that
forms the ceiling on sustainable levels. Cordella and Levy Yeyati (2005) argue that a ceiling 8 Cordella and Levy Yeyati (2005) suggest “smoothing” the eligibility criteria so that exogenous shocks temporarily pushing a country about the eligibility threshold do not precipitate a sudden jump in interest rates and/or a crisis. While this might help for small exogenous shocks that the authorities wish to offset, it will not help for large exogenous shocks that cannot be offset in the medium term.
- 11 -
should be defined for the overall debt-GDP ratio but also recommend that foreign-currency
debt and short-term debt should receive heavier weights in the calculation of this total. They
suggest imposing a ceiling on the fiscal deficit in each of the preceding three years.
Unfortunately, in the real world sustainability depends on forecasts of future growth
rates and interest rates that are disputable and uncertain. It depends on estimates of the
political will of a government and society to mobilize and transfer real resources for purposes
of debt service. Given this uncertainty, it seems unavoidable that any insurance facility will
occasionally lend to countries that find it impossible to repay. Or it will not lend to countries
whose problems are liquidity related, leading to complaints and recrimination. Any
automatic scheme that depends on the existence of an operational distinction between
insolvent and illiquid crisis countries is unlikely to be feasible in practice.9
3. The Sample and its Properties
We analyze data since 1980 on all developing countries for which the relevant data
are available, 24 emerging markets and 59 other economies. Emerging markets are defined
as countries included in the Morgan Stanley Capital International (MSCI) index. All other
countries are labeled as developing.10 Following Calvo et al. (2004) and Cavallo and Frankel
(2004), we consider both the first and the second moments of the capital flow series in
9 This argument is developed and defended at greater length in Eichengreen (2002).
10 We started with the universe of countries but dropped those with less than 1 million population, major oil producing countries, as well as industrial countries which are usually net exporters of capital rather than net importers of capital. Our final sample was based on the availability of data in the remaining group of countries
- 12 -
identifying sudden stops,. As our measure of capital flows we use the financial account
balance, which is a summary measure of net capital flows and includes net foreign direct
investment, net portfolio investment and other investments. First, we identify the years
where the financial account balance exhibited a large decline relative to its long term
average. We require this to be a large discrete drop and not just a correction of a large
temporary inflow. Second, we calculate the mean and one-standard deviation band of the
financial account balance using the data up to the years identified in the first stage as
potential sudden stops and retain only years that qualify on the basis of this criterion.11 Other
authors have used two-standard deviation bands, but because we are using annual data, a two
standard deviation criterion turns out to be very strict.
In practice, our dates are very close to those of Calvo et al (2004) and Cavallo and
Frankel (2004). On the other hand, they differ from those of Edwards (2005), where a
sudden stop is said to occur when a country had previously been receiving significant capital
inflows (where it ranked in its region’s third quartile in terms of capital inflows in the two
previous years) and their volume declines by at least 5 per cent of GDP in a given year. This
is a lenient criterion that tends to capture episodes where the financial account balance
declines only from, say, 20 per cent to 15 per cent of GDP, as well as episodes where there is
a correction of a temporary increase in the balance the year before.
Figure 1 shows the time profile of these events. Clearly there is bunching of sudden
stops, as emphasized by Calvo (2005). Peaks coincide with the 1982 debt crisis, the Asia-
11 The moments were calculated using data up to the crisis year and not including the potential crisis year.
- 13 -
Russia-LTCM crisis of 1997, and the aftermath of the Argentine crisis in 2002. Typically,
these episodes end relatively quickly (another point emphasized by Calvo). The majority (58
per cent) last only one year, and more than three quarters (77 per cent) are resolved within
two years.12
Typically, countries experience a net capital outflow on the order of 6 per cent of
GDP in the first two years of a sudden stop episode (Figure 2). (Here and in what follows,
the broken line is for emerging markets while the solid line for all developing countries.) By
this standard, the portfolio outflow is relatively small, on the order of ¼ percent of GDP.
There is no discernible impact on net FDI flows. By implication, net capital outflows consist
mostly of other forms of capital, such funds channeled through the banking system,
commercial credits, and so forth. (See also Table 3.)
We can similarly examine the behavior of ancillary variables in these episodes.
Typically, countries experiencing a sudden stop also experience a deterioration in the current
account balance on the order of 3 per cent of GDP. (See Figure 3.) Note that we are not
looking here at the incidence of “current account reversals,” where a current account reversal
is typically defined as an episode when there is a large reduction in the size of an outstanding
current account deficit; rather, we are summarizing the average behavior of the current
account balance in our episodes of capital account reversal. Predictably, larger current
account reversals are evident in countries in which the sudden stop lasts more than 2 years.
The current account being the difference between savings and investment, the consequences
for the current account are necessarily reflected in these variables. Figure 3 shows that the 12 12 per cent last three years and an additional 12 per cent last still longer than that.
- 14 -
action is mainly on the investment side – capital formation declines sharply, especially in
emerging markets. It follows that GDP growth is negative on average, although it rebounds
after one year (Figure 3). A statistical summary of these patterns is in Table 4.
Sudden stops are also not the same as currency crises. For emerging markets we use
crisis dates from Kaminsky and Reinhart (1999) and Berg and Patillo (1999); for developing
countries we use the dates in Frankel and Rose (1996) and Milesi-Ferreti and Razin (1998).13
Only about 15 per cent of the sudden stops in our sample are associated with
contemporaneous currency crises (see the first panel of Table 5). The picture looks the same
when we lag the currency crises (as in the second panel of Table 5). But the probability of a
sudden stop rises from 6 per cent to 11 per cent if there is a currency crisis the year before.
While we have fewer observations for banking crises, the pattern here appears to be
different. On the one hand, a large share of sudden stops (35 percent) appears to coincide
with banking crises (see the third panel of Table 5). But there is only a small increase in the
probability of a sudden stop if there is a banking crisis in the preceding year, from 6 to 7 per
cent (fourth panel of Table 5). This suggests a complex relationship between sudden stops,
currency crises and banking crises. (See also Table 6.) It does not suggest that causation
runs exclusively from banking crises to currency crises (compare Kaminsky and Reinhart
1999).
13 For recent years we supplement these with the dates from Frankel and Wei (2004). These data are merged following the procedures in Gupta, Mishra and Sahay (2003).
- 15 -
4. Basic Results
Information on IMF programs is drawn from a data base maintained by the Fund’s
Policy Development and Review Department. It includes information on the year the
program started and ended, whether the program was precautionary from the outset or turned
precautionary in the course of its operation, the value of the funds approved, and amount
used under the program. Under precautionary programs, the IMF and the country agree on
conditionality and monitoring but the country has no intention of drawing resources from the
Fund. A quarter of the programs we consider are in emerging markets. Slightly more than
10 per cent of the Fund programs in our sample were precautionary from the start, 12 percent
in the case of emerging markets.
Table 6 is a first look at the association between Fund programs and sudden stops.
For simplicity we concentrate on programs that started in the same year as the sudden stop.
In approximately 30 percent of sudden stop cases, a new IMF program is negotiated in the
same year. The pattern is even more prevalent in emerging market economies, where over
42 percent of sudden stops coincide with the negotiation of an IMF program.
To be sure, this tells us nothing about causation. Lagging the data on new programs
is one way of increasing the likelihood that what we are picking up is causality running from
the program to the likelihood of a sudden stop.14 Whereas the unconditional probability of a
sudden stop is 5.6 per cent, the probability of a sudden stop in the year after an IMF program
is 4.9 per cent. This last figure is essentially the same when we instead consider whether a
country had a Fund program in place in either of the two preceding years. Thus, to the extent 14 Below we attempt to identify this effect more precisely using instrumental variables.
- 16 -
that the probability of a sudden stop declines with the negotiation of an IMF program, the
difference is small.15
The raw data are consistent with the idea that whether or not a Fund program is in
place matters for the behavior of consumption, the trade balance and the current account in
the wake of the sudden stop.16 Here we are looking at all programs, whether initiated in the
two years immediately preceding the crisis or prior to that. Program countries indicated by
the broken line receive credit from the IMF (Figure 4) that enables them, not surprisingly, to
run larger current account deficits and maintain higher levels of consumption, even though
the shift in their financial accounts (our measure of the severity of the sudden stop itself) is
no different than in non-program countries. The impact on the other variables is unclear.
There is some sign that program countries enjoy more stable portfolio capital flows and that
growth recovers more quickly following the sudden stop, although the latter may simply
reflect the rubber band effect associated with the faster decline in output before the fact.
15 Unfortunately, we have only two cases where a precautionary program preceded a sudden stop: Argentina in 2001 and Brazil in 2002. For what they are worth, these data suggest that the probability of a sudden stop in the year after a precautionary IMF program is 4.7% (3.3 for precautionary program in either of the two preceding years). Note that we can still attempt to distinguish the effects of precautionary programs in econometric analysis below, since we also have instances of such programs in country-year cases where there were no sudden stops.
16 Here we are not lagging whether there is a Fund program in place by one year. It should be easy enough for the reading to make this visual adjustment.
- 17 -
5. Multivariate Analysis
We now present a regression analysis of the association of IMF programs with
sudden stops, conditioning on other determinants of the change in the financial accounts and
instrumenting for the endogeneity of the program variable. The first set of regressions looks
at the determinants of sudden stops. Our controls, drawn from the earlier literature, include
U.S. interest rates and growth rates as measures of global economic conditions, the subject
country’s own lagged GDP and export growth rates as measures of domestic conditions, and
the debt service/export and domestic credit/GDP ratios, also lagged, as measures of the
domestic financial situation. We construct the dependent variable two ways: first, as a
simple binary indicator equaling one in the first year of a sudden stop; second, as the
financial account balance if and only if a sudden stop occurred as a way of capturing the
severity of the event.17 Regressions using the first dependent variable are estimated by probit
(we report the marginal probabilities), while regressions using the second dependent variable
are estimated by tobit. In principle, the tobit regressions should make more complete use of
the available information. Country and in some specifications time dummies are included.
Benchmark regressions are in Table 9. Faster domestic growth in the preceding
period appears to reduce the likelihood of a sudden stop, while more external debt and
greater debt servicing obligations increase the likelihood of this event. There is evidence that
a higher domestic credit ratio also increases the likelihood of a sudden stop, but only in the
probit regressions. On the other hand, the incidence of sudden stops is not significantly 17 Instead of financial account balance, we also look at the change in financial account balance in the year of the sudden stop as compared to the average of pre sudden stop years. The two measures are highly correlated, and the results are similar across the two measures.
- 18 -
affected by measures of global conditions (the U.S. growth rate and interest rate).18
Experimentation with a variety of additional measures of global conditions confirmed this
result; we therefore dropped these measures in the subsequent analysis. In addition, dropping
the country dummies and thus adding cross section variation leaves the results basically
unchanged.
Table 10 adds a number of additional variables to this benchmark specification.
Countries with fixed exchange rates, the alternative omitted from the table, appear to be more
susceptible to sudden stops.19 We also find that countries better integrated into international
capital markets are less vulnerable to sudden stops.20 This may reflect the fact that countries
better integrated into international financial markets have stronger institutions.
6. IMF Programs and Sudden Stops
Table 12 adds the IMF program variables. When we measure this by the existence of
a program, there is little effect. But when we instead consider the initiation of a program in
the same or immediately preceding year, the coefficient in question is negative and
significantly different from zero. This pattern of results is difficult to reconcile with the
insurance analogy; if IMF credit provides insurance against sudden stops, it is not obvious
18 Contrary to the emphasis in Calvo et al. (2004).
19 We use the Reinhart and Rogoff 2003 measure of de facto exchange rate arrangements.
20 Note that this is the opposite of what Edwards (2005) finds for emerging markets, although he uses a different measure of financial openness. This difference in means also seems to be heavily driven by Asian countries. We use the Mody and Murshid 2005 measure of financial integration.
- 19 -
that a long-standing credit line should be less effective than a newly-authorized one, other
things equal. An alternative interpretation is that we are picking up a signaling effect: that
the government’s willingness to agree to a set of conditions with the IMF provides
reassurance to investors. This is consistent with previous results reported by Marchesi
(2001).
The insurance analogy is sometimes taken as an argument for exceptional access – as
justifying a credit line sufficiently large to reassure private investors. This suggests
distinguishing effects by the size of the program. We do so in two ways, scaling programs
alternatively by external debt and GDP, lagging both ratios by one period. Both variables
enter negatively, as predicted, but only IMF credit as a share of GDP approaches significance
at the 90 per cent level, and only when it is the sole measure of the program included in the
equation. Again, this is somewhat difficult to reconcile with the insurance interpretation.21
A key question is whether the negative association between programs and sudden
stops reflects causality running from the latter to the former. It is not clear in this context
whether endogeneity is a serious problem; one can argue that reverse causality should bias
the coefficient toward zero, since a sudden stop will increase the likelihood of a Fund
program and thus produce a positive relationship, where we find a negative coefficient.22 Be
this as it may, the appropriate treatment for this problem is instrumental variables. We
21 More experimentation with alternative scale variables – for instance, scaling the size of the program by the average value of the financial-account balance in preceding years – might produce more positive findings, but this would also open one up to accusations of data mining.
22 We provide some evidence on this below.
- 20 -
follow the strategy of Barro and Lee (2004) for instrumenting IMF programs. These authors
take a political economy approach to modeling the incidence of programs, arguing that
decision making in the Fund is influenced by the organization’s principal shareholders, above
all the United States. They model the likelihood of a country negotiating a program as a
function not just of country characteristics but also of its links with the United States. To
capture the latter, they include U.S. aid as a percentage of total foreign aid received by the
country and the share of votes in the UN General Assembly in which the country voted the
same way as the United States, both lagged.23 We also include reserves to imports, and lag
each variable by two years. We estimate two first-stage specifications, first with one lag of
real GDP growth, exports growth, domestic credit, debt servicing, and external debt; and
another with first and second lags of each of these variables.
Table 12 shows the first- and second-stage results. In the first-stage, both U.S. aid
and UN voting patterns enter with their predicted positive signs, although only the aid
measure is significantly different from zero.24 The predicted probability of a new IMF
23 The authors also use past participation in Fund programs as an instrument. The problem with using lags of the endogenous variable as an instrument is of course that it may be picking up omitted country characteristics that are slow to change and durably associated with financial problems; see Mody and Stone (2005). Alternatively, we can follow Celasun and Ramcharan (2005) by using the share of G3 exports going to each subject country as a measure of the importance the principal shareholders may attach to extending assistance through the Fund. However this variable is a poor instrument for the incidence of the IMF program. The coefficient of this variable is consistently negative and significant in the first stage regressions of IMF program. One interpretation is that it is a proxy for the size of the country and the larger countries are probably less susceptible to problems necessitating an IMF program.
24 Use of voting in the UN General Assembly has been criticized in the political-science literature for being dominated by non-consequential votes and thus containing relatively little
(continued)
- 21 -
program based on this regression continues to enter in the second stage. Both the level of
statistical significance and the absolute value of the effect are greater than before. The fact
that we now get a larger negative coefficient for the effect of a new Fund program on the
likelihood of a sudden stop is consistent with the notion that there also exists reverse
causality making for a positive correlation and biasing the OLS estimates toward zero.
7. Robustness Checks and Extensions
We next considered a variety of robustness checks and extensions. These produced
mixed results—which is another reason not to claim too much for the findings. On the
positive side, subsample estimates suggest that the relationship between sudden stops and
Fund programs is stable over time. For example, we get the same result when we drop
observations for the 1980s from the sample.
On the negative side, tests of whether IMF programs had a more powerful preemptive
effect on sudden stops for countries with strong fundamentals, as suggested by the insurance
analogy, failed to turn up positive evidence. We distinguished countries by the level of
external debt and debt-service payments. Indebtedness is labeled high if debt as a percentage
of GDP exceeds 70 percent; medium if it is between 40 and 70 percent of GDP (roughly top
1/3 and middle 1/3 of observations respectively in the sample). Debt servicing is labeled high
if debt servicing as a percentage of exports exceeds 27 percent and medium if it is between
13 and 27 percent (roughly top 1/3 and middle 1/3 of observations in the sample). In the last
information on political affinity. This is one interpretation of what we find in the first-stage regressions.
- 22 -
two rows debt servicing and debt are considered to be jointly high if both debt and debt
servicing are high and considered to be jointly medium if they are both neither high nor low.
Here low is the omitted alternative. While the coefficients on medium and high indebtedness
are positive, as anticipated, they do not differ significantly from zero.25
Finally, in addition to the impact of IMF programs on the incidence of sudden stops,
we analyzed their impact on the associated output losses. In the regressions reported in
Table 14, the dependent variable is GDP growth (at constant prices). The control variables
are lagged per capita income in U.S. dollars (expressed in logs), lagged economic growth and
time fixed effects. To identify the impact of Fund programs on growth during sudden stops,
in the first two columns we include the probability of an IMF program (the first-stage
regression is that in the first column of Table 12), a dummy for sudden stop incidents (not the
first year but the entire duration), and an interaction term for program and sudden stop. We
estimate this specification with and with out country fixed effects. The basic results offer
little support for the hypothesis that Fund programs significantly attenuate the output effects
of sudden stops.
Of course, our sudden-stop cases are not drawn randomly from the larger sample of
observations. In the remaining columns of Table 14 we therefore provide some simple
treatments for the associated selectivity. We use the specification in Table 12 (column 1) to
25 In earlier work, analyzing the “catalytic effect” of the IMF on capital flows, we do find evidence of threshold effects, especially for access to bond markets (Mody and Sarvia, 2003 and Eichengreen, Mody, and Kletzer, 2005). In particular, countries in an intermediate state of vulnerability seemed to benefit from improved market access while others did not. It could be that these threshold effects do not operate in extreme distress situations. We expect to pursue this line of enquiry further.
- 23 -
estimate the probability of a sudden stop and use this to construct the Inverse Mills Ratio.
We then add this as an additional control to our basic growth regressions. Now we get a
positive estimate of the IMF program, though it is not statistically significant. We also
estimate these equations for the subsample of emerging markets in the period since 1990.
Here too there is some evidence of positive effects of IMF programs on growth.26
8. Conclusions and Implications
In this paper we have presented the first systematic attempt to identify the impact of
IMF programs on sudden stops. The literature on self-fulfilling crises and the insurance
motive for IMF lending suggests that Fund programs could reduce the incidence of these
episodes characterized by disruptive reversals of capital flows. In contrast, the large
literature critical of the Fund suggests that its programs, especially those negotiated with
emerging markets in the 1990s, tend to be too small, too late, and too laden with burdensome
economic and political conditions to effectively restore confidence. Thus, the large rescue
packages negotiated with Thailand and other Asian countries in 1997 did not prevent painful
and disruptive sudden stops. In other cases like Brazil in 2002, the sudden stop was not
averted, but negative output effects seem to have been minimized.27
While these are prominent cases, it is still not clear how representative they are of
general experience. In the case of Thailand, it can be argued that the problem was that the
26 Adding country dummies to the specifications with correction for selection bias gave unstable results and so are not reported here.
27 Reviews of this experience include Ghosh et al. (2002) and Independent Evaluation Office (2003).
- 24 -
program was not in place prior to the sudden stop and that by the time it was negotiated
things had gotten out of hand. In the case of Brazil in 2002, it might be argued that the fact
that the world economy was recovering was important for minimizing the output effects of
sudden stops. As in these examples, much of the evidence invoked in this debate is
anecdotal. A more systematic analysis is clearly desirable.
Overall, our results are consistent with the notion that IMF programs have some
positive effective in reducing the incidence of sudden stops. At the same time, the evidence
to this effect comes packaged with surprises and anomalies. There is little evidence that
larger Fund programs more effectively inoculate countries against sudden stops. Newly-
negotiated programs seem to be more effective in this regard than long-standing
arrangements. It is tempting to interpret both observations as indicating that the signaling
effect of IMF programs matters more than the emergency financial assistance. We are
unable to identify evidence that IMF programs are more effective at insulating countries from
sudden stops when they already have fundamentally strong policies in place. Turning to
output effects, we find some evidence that the presence of a Fund program reduces the
negative impact of the sudden stop on growth. However, this result depends on correcting
for the selectivity associated with the sudden stop episodes.
The literature on the effects of IMF programs is notorious for its methodological
limitations. It is a fact of international financial life that countries that approach the Fund
differ systematically from other countries, creating problems of endogeneity and selectivity.
In this paper we have described some instrumental-variables strategies and statistical
adjustments for these problems. However, these approaches have limitations, and many of
- 25 -
the same critiques levied against the related literature apply to the results reported here.
These are reasons for regarding the results as a first cut and not a definitive guide to policy.
Even if one accepts that there is evidence that IMF programs reduce the incidence and
virulence of sudden stops, this leaves open the question of whether and how the Fund’s
lending practices should be adapted in response to these findings. One view would be that
such findings strengthen the case for a generously endowed, quick disbursing, automatic
facility for which countries with strong policies would presumably be prequalified.28 A quite
different view would be that prequalification is impractical, that automaticity is infeasible,
and that larger credit lines implying the provision of additional resources for the Fund are not
in the cards. We trust that these policy implications will be hotly debated at the Rio
conference, but they are not the subject of this paper.
28 Again, for some cautious arguments along these lines, see Ostry and Zettelmeyer (2005).
- 26 -
Table 1. Literature on Sudden Stops Authors/Year Countries/Types/Years Dependent Variable Main Explanatory Variables
Sudden stop dummy Calvo, Izquierdo, and Mejia (2004)
1) Degree of domestic liability dollarization
15 emerging market countries and 17 developed countries, 1990-2001
Doaresto
2) 1 minus (the degree of trade openness) -- to proxy for the sensitivity of the real exchange rate to capital flow reversals.
ThetheHigpardeg
Control variables: two measures of exchange rate regime flexibility, ratio of foreign reserves to current account deficit, credit growth, FDI, fiscal balance, terms of trade growth, public sector debt, yearly dummies.
Frankel and Cavallo (2004)
141 countries, 1970-2002 Sudden stop dummy 1) Trade openness
2) Foreign debt to GDP 3) Current account balance to GDP
Opstocorgra
Sudacc
Control variables: log of reserves in months of imports, log of GDP per capita, FDI/GDP, institutional quality ratio of short term debt to total debt, and a measure of nominal exchange rate rigidity For
sig
Hutchison and Noy (2004) GDP growth rate 1) Currency crisis dummy
27 emerging market countries, 1975-1997 2) Current account reversal dummy
3) Sudden stop crisis dummy (interaction of the two dummy variables above)
Cusigstoimpcur
Control variables: (lagged) real GDP growth, change in budget surplus to GDP ratio, credit growth, foreign output growth, real exchange rate overvaluation, openness, country dummies.
Edwards (2005) 157 countries, 1970-2000 GDP Per capita 1) Sudden stop dummy
2) Current account reversal dummy
Control variables: External terms of trade shocks (change in terms of trade), the growth gap (the difference between the long run rate of real per capita GDP and the current rate)
Revnegbotincdumthe
- 27 -
Table 2. Literature on IMF Programs Authors/Year Countries/Types/Years Dependent Variable(s) Measurement
Goldstein and Montiel (1986) 58 developing countries 1) Overall balance of payments/ GNP 1974-1981 2) Current account/GNP
3) Inflation rate
The mean change betprogram periods acroscountries for each of t
4) GDP growth
The difference in meavariables for program
The sign and statisticacoefficients of the IMF
Pastor (1987) 18 Latin American countries 1) Balance of payments measures
1965-1981 2) Inflation rate 3) GDP growth
The mean change betprogram periods acroscountries for each of t
4) Labor's share of income
Khan (1990) 69 developing countries 1) Balance of payments to GDP 1973-1988 2) Current account surplus to GDP 3) Inflation rate
The mean change betprogram periods acroscountries for each of t
4) Real GDP growth rate
The difference in meavariables for program
The sign and statisticacoefficients of the IMF
- 28 -
Authors/Year Countries/Types/Years Dependent Variable(s) Measurement
Conway (1994) 74 developing countries The sign and statisticacoefficients of:
1976-1986 1) A binary indicator o
1) Economic performance -- GDP growth, domestic investment to GNP ratio, current account-to-GNP ratio, inflation rate.
2) Percentage of year
3) Percentage of availperiod t that is disburs
2) Policy variables -- the ratio of government consumption to GNP, the rate of domestic credit creation, budget surplus/deficit as a share of GNP, real exchange rate.
74 low-income countries 1) Real GDP growth rate Dicks-Mireaux, Mecagni, and Schadler (2000) 1986-1991 2) Consumer price inflation
The sign and statisticacoefficients of the IMF
3) The ratio of external debt/service to exports (measure of external viability)
79 developing countries Output growth Przeworski and Vreeland (2000) 1971-1990
The difference betweein the two situations (uprograms) independen
- 29 -
Authors/Year Countries/Types/Years Dependent Variable(s) Measurement
Hutchison (2001) 67 developing countries Real GDP growth The sign and statisticacoefficients of:
1970-1997 1) IMF program dumm
2) Interaction term betdummy and the IMF padditional effects on ocurrency crisis that is IMF program)
Barro and Lee (2002) 81 developing countries GDP growth rate The sign and statisticacoefficients of:
1975-1999
1) IMF loan size -- theIMF loans to GDP
2) IMF participation --each five-year period tloan program
Evrensel (2002) 91 developing countries 1) Current account 1971-1997 2) Overall balance of payments 3) Reserves
The mean change betprogram periods acroscountries for each of t
4) Domestic credit 5) Inflation rate 6) Budget deficit 7) Net domestic borrowing 8) Net foreign borrowing 9) Real growth rate
- 30 -
Authors/Year Countries/Types/Years Dependent Variable(s) Measurement
Hardoy (2002) 109 developing countries
1970-1990
The annual rate of growth of real GDP per capita
Matching estimator = avalues of the outcomeobservations and the mobservations.
Difference-in-differencdifference between theprogram and post-prog
Hutchison and Noy (2003) 67 developing countries Real GDP growth The sign and statisticacoefficients of:
1975-1997 1) IMF program dumm
2) Latin American IMF 3) IMF program dumm 4) Latin American IMF
29 emerging market countries 1) Current account balance to GDP Bordo, Mody, and Oomes (2004)
1980-2002 2) Reserves to imports 3) Short-term debt to reserves
Difference between prnonprogram participatfundamentals) in probstate
- 31 -
Authors/Year Countries/Types/Years Dependent Variable(s) Measurement
Atoyan and Conway (2005) 95 developing countries Difference in intercept
1993-2002
1) Changes in real per capita economic growth rate
2) Changes in fiscal surplus to GDP ratio 3) Changes in current account surplus to GDP
ratio Coefficient of the propof probability of partici
Intercept coefficient
Dreher (2005) 98 developing countries The sign and statisticacoefficients of:
1970-2000 (5-year intervals)
Average five-year growth rate of per capita GDP
1) IMF programs -- comeasuring the fractionoperates under an IMF
2) IMF loans to GDP
3) Compliance with co
i) No program suspe2001)
ii) Equally phased lo2003)
iii) The share of the disbursed (Killick 1995
- 32 -
Figure 1. Year-Wise Probability of a Sudden Stop
Probability of a Sudden Stop
0
5
10
15
2019
8119
82
1983
1984
1985
1986
1987
1988
1989
1990
1991
1992
1993
1994
1995
1996
1997
1998
1999
2000
2001
2002
Probability of a sudden stop_all countries probablity of a sudden stop_emerging markets
- 33 -
Figure 2. Magnitude and Composition of Financial Flows
Financial Account Balance (as % of GDP)
-8-6-4-20
246
8
-3 -2 -1 0 1 2 3
All Countries Emerging Markets
Net FDI Flows (as % of GDP)
0
0.5
1
1.5
2
-3 -2 -1 0 1 2 3All Countries Emerging Markets
Net Portfolio Flows (as % of GDP)
-0.5
0
0.5
1
1.5
2
-3 -2 -1 0 1 2 3
All Countries Emerging Markets
Net Other Flows (as % of GDP)
-8
-6
-4
-2
0
2
4
6
-3 -2 -1 0 1 2 3
All Countries Emerging Markets
- 34 -
Table 3. Composition of Capital Flows (all expressed as percentage of GDP) During and Prior to Sudden Stops
Average of Three Years Prior
to Sudden Stops During Sudden Stops
(first year) Total net financial assets 2.2
[3.8] -4.7 [4.8]
Net FDI 1.1 [1.6]
.96 [1.8]
Net portfolio flows .34 [1.1]
-.16 [1.0]
.84 [3.5]
-5.6 [5.4]
.66 [2.2]
-1.9 [5.1]
.13 [1.43]
-1.7 [2.7]
Net other capital flows Net other capital flows_government. Net other capital flows_banks Net other capital flows_other
.05 [2.1]
-1.7 [3.2]
Note: Standard deviations are given in brackets.
- 35 -
Figure 3. Macro Economic Effects of Sudden Stops
Current Account Balance (as % of GDP)
-7-6-5-4-3-2-1012
-3 -2 -1 0 1 2 3
All Countries Emerging Markets
Gross Capital Formation (as % of GDP)
17
19
21
23
25
27
29
-3 -2 -1 0 1 2 3
All Countries Emerging Markets
Real Growth of GDP
-1
0
1
2
3
4
5
6
-3 -2 -1 0 1 2 3
All Countries Emerging Markets
- 36 -
Table 4. Real Effects of Sudden Stops During and Prior to Sudden Stops Average of Three Years
Prior to Sudden Stops
During Sudden Stops Current account balance_GDP
-4.4 [5.1]
-2.0 [7.0]
Real growth of GDP 2.7 [3.1]
1.3 [5.1]
Export growth .34 [2.6]
.47 [2.8]
Capital formation_GDP 22.7 [7.8]
20.3 [7.7]
Fixed capital formation_GDP
21.6 [7.3]
19.4 [7.3]
Private consumption growth 2.2 [3.9]
-.42 [9.9]
Table 5. Sudden Stops and Banking and Currency Crises
A. Sudden Stops (SS) and Currency Crises (CC)
No SS SS Total No CC 1250 74 1324 CC 105 15 120 Total 1355 89 1444
B. Sudden Stops (SS) and Lagged Currency Crises (CC) No SS SS Total No CC (Lag) 1239 74 1313 CC (Lag) 116 14 130 Total 1355 88 1443
C. Sudden Stops (SS) and Banking Crises (BC) No SS SS Total No BC 981 50 1031 BC 241 27 268 Total 1222 77 1299
D. Sudden Stops (SS) and Lagged Banking Crises (BC) No SS SS Total No BC (Lag) 963 56 1019 BC (Lag) 239 19 258 Total 1202 75 1277
- 38 -
Table 6. Currency Crises, Banking Crises, Twin Crises and Sudden Stops
Number of Respective
Crises Number of
Sudden Stops
Conditional Probability of a Sudden Stop (conditional on
respective crises) Currency crises 133 18 13.5 Banking crises 268 27 10.1 Twin crisis 35 8 22.9 Currency crises_first year 120 15 12.5 Banking crises_first year 97 16 16.5 Twin crisis_first year 14 5 35.7
Total number of
observationsNumber of
sudden stops Unconditional probability of a
sudden stop All observations 1880 104 5.5
- 39 -
Table 7. IMF Programs in the Sample
A. IMF Programs, Precautionary and Other, Full Sample
Precautionary program from the start IMF program 0 1 Total 0 1541 0 1541 1 391 46 437 Total 1932 46 1978
B. IMF Programs, Precautionary and Other, Emerging Markets
Precautionary program from the start IMF program 0 1 Total 0 470 0 470 1 95 13 108 Total 565 13 578
- 40 -
Table 8. IMF Programs and Sudden Stops
A. IMF Programs and Sudden Stops in all Countries
IMF Program Sudden stops 0 1 Total 0 1398 378 1776 1 74 30 104 Total 1472 408 1880
B. IMF Programs and Sudden Stops in Emerging Countries IMF Program Sudden stops 0 1 Total 0 423 84 507 1 22 16 38 Total 445 100 545
C. Lagged IMF Programs and Sudden Stops
Lagged IMF Program—One
Year Lag Sudden stops 0 1 Total 0 1322 372 1694 1 85 19 104 Total 1407 391 1798
D. Lagged Precautionary Programs and Sudden Stops
Lagged IMF Program—Up to Two Years
Sudden stops 0 1 Total 0 1658 40 1698 1 102 2 104 Total 1760 42 1802
- 41 -
Figure 4. IMF Programs and Effects of Sudden Stops
`
Financial Account Balance (as % of GDP)
-8
-6
-4
-2
0
2
4
6
-3 -2 -1 0 1 2 3
All Countries IMF program
Real Growth
0
1
2
3
4
5
-3 -2 -1 0 1 2 3
All Countries IMF program
Current Account Balance (as % of GDP)
-9-8-7-6-5-4-3-2-10
-3 -2 -1 0 1 2 3
All Countries IMF program
Net Portoflio Flows (as % of GDP)
-0.4
-0.2
0
0.2
0.4
0.6
0.8
-3 -2 -1 0 1 2 3
All Countries IMF program
Capital Formation (as % of GDP)
15
20
25
-3 -2 -1 0 1 2 3
All Countries IMF program
Net Other Flows (as % of GDP)
-8
-6
-4
-2
0
2
4
6
-3 -2 -1 0 1 2 3
All Countries IMF program
- 42 -
Table 9. Determinants of Sudden Stops and Their Intensity
Columns reporting probit results present marginal probabilities, based on the STATA command “dprobit”. Robust z statistics are presented in parentheses. *** represents significance at 1 percent, ** at 5 percent, and * at 10 percent.
Dependent Variable Indicator for First Year of Sudden Stop (Probit)
Financial Account Outflow if a Sudden Stop Occurred (Tobit)
US treasury bill rate -0.004 [1.39]
-0.004 [1.43]
-0.02 [1.12]
-0.02 [1.12]
US real growth rate 0.13 [0.36]
0.18 [0.46]
-0.29 [0.12]
-0.33 [0.15]
GDP Growth, lag -0.002 [1.45]
-0.002 [1.35]
-0.002 [1.45]
-0.003* [1.65]
-0.02** [2.00]
-0.02* [1.92]
-0.02** [2.00]
-0.02* [1.70]
Export growth, lag -0.49 [1.50]
-0.48 [1.49]
-0.49 [1.50]
-0.52* [1.73]
-4.69*** [2.71]
-4.59*** [2.66]
-4.68*** [2.71]
5.31*** [2.96]
External debt/GDP, lag 0.04*** [2.73]
0.05*** [2.99]
0.04*** [2.74]
0.04*** [2.87]
0.09 [0.89]
0.11 [1.13]
0.09 [0.89]
0.10 [1.00]
Debt servicing /Exports, lag 0.12* [1.91]
0.104 [1.63]
0.12* [1.88]
0.11* [1.83]
1.69*** [4.68]
1.65*** [4.58]
1.69*** [4.67]
1.80*** [4.75]
Domestic Credit /GDP, lag 0.15*** [3.01]
0.16*** [3.21]
0.15*** [3.01]
0.15*** [3.39]
0.002 [0.01]
0.04 [0.15]
0.003 [0.01]
0.04 [0.13]
Country fixed effects yes yes yes yes yes yes yes yes Time fixed effects no no no yes no no no yes Observations 1118 1118 1118 1118 1401 1401 1401 1401
- 43 -
Table 10. Financial Mobility, Exchange Rate Regimes, and Sudden Stops
Columns reporting probit results present marginal probabilities, based on the STATA command “dprobit”. Robust z statistics are presented in parentheses. *** represents significance at 1 percent, ** at 5 percent, and * at 10 percent.
Dependent Variable Indicator for First Year of Sudden Stop (Probit)
Financial Account Outflow if a Sudden Stop Occurred (Tobit)
GDP growth, lag -0.003 [1.61]
-0.001 [0.34]
-0.001 [0.50]
-0.02 [1.53]
-0.01 [0.80]
-0.01 [1.03]
Export growth, lag -0.47 [1.55]
-0.41 [1.48]
-0.36 [1.34]
-5.2*** [2.91]
-5.4*** [2.89]
-4.6** [2.55]
Debt servicing/exports, lag 0.11* [1.88]
0.23*** [4.04]
0.206*** [3.52]
1.80*** [4.74]
2.41*** [5.61]
2.46*** [5.88]
Domestic credit/GDP, lag 0.14*** [3.40]
0.16*** [3.28]
0.164*** [3.66]
0.05 [0.18]
0.17 [0.57]
0.24 [0.82]
External debt/GDP, lag 0.04** [2.43]
0.05** [2.38]
0.052*** [2.65]
0.06 [0.59]
0.06 [0.54]
0.05 [0.41]
Financial mobility lag -0.01* [1.65]
-0.01 [1.44]
-0.005 [0.53]
-0.16*** [2.92]
-0.17*** [2.83]
-0.14** [2.35]
ER regime: limited flexibility -0.05** [2.35]
-0.02 [0.10]
ER regime: float -0.04** [2.20]
-0.35* [1.89]
ER regime: free fall 0.03 [0.79]
0.35* [1.66]
ER regime: limited flexibility, lag -0.037 [1.60]
-0.06 [0.34]
ER regime: float, lag -0.019 [0.86]
-0.15 [0.83]
ER regime: free fall, lag 0.028 [0.88]
0.27 [1.31]
Country fixed effects yes yes yes yes yes yes Time fixed effects yes yes yes yes yes yes Observations 1118 843 1401 1139 1192
- 44 -
Table 11a. IMF Program and Probability of Sudden Stops (Probit Estimates)
Columns reporting probit results present marginal probabilities, based on the STATA command “dprobit”. Robust z statistics are presented in parentheses. *** represents significance at 1 percent, ** at 5 percent, and * at 10 percent.
Dependent Variable Indicator for First Year of Sudden Stop (Probit) GDP growth, lag -0.003*
[1.67] -0.003* [1.73]
-0.003* [1.74]
-0.002 [1.47]
-0.002 [1.44]
-0.003* [1.67]
-0.003 [1.57]
-0.001 [0.27]
Export growth, lag -0.51* [1.68]
-0.51 [1.64]
-0.50 [1.62]
-0.56* [1.79]
-0.57* [1.82]
-0.49 [1.59]
-0.53* [1.75]
-0.37 [1.37]
Debt servicing/exports, lag
0.11* [1.85]
0.12** [2.03]
0.13** [2.13]
0.08 [1.30]
0.08 [1.39]
0.13** [2.12]
0.14** [2.49]
0.25*** [4.34]
Domestic credit/GDP, lag
0.15*** [3.39]
0.14*** [3.28]
0.14*** [3.21]
0.14*** [3.13]
0.14*** [3.21]
0.14*** [3.31]
0.15*** [3.25]
0.14*** [2.97]
External debt/GDP, lag
0.04*** [2.90]
0.04*** [2.74]
0.04*** [2.70]
0.04*** [2.69]
0.04*** [2.67]
0.04*** [2.81]
0.06*** [3.04]
0.04** [2.17]
IMF program existed in T, T-1
0.01 [0.66]
New IMF program in T, T-1
-0.03** [2.45]
-0.03** [2.26]
-0.04** [2.06]
-0.04** [2.29]
-0.03** [1.99]
-0.03* [1.88]
-0.04** [2.88]
Precautionary program in T,T-1
-0.04 [1.17]
Approved amount/GDP, lag
0.002 [0.47]
Approved amount/debt, lag
0.001 [0.54]
IMF credit/debt, lag -0.28 [1.10]
IMF credit/GDP, lag -0.01** [2.10]
-0.002 [0.44]
Financial mobility lag -0.02* [1.95]
ER regime: limited flexibility
-0.06** [2.55]
ER regime: managed float
-0.05** [2.33]
ER regime: free fall 0.03 Country Fixed Effects Yes Yes Yes Yes Yes Yes Yes Yes Year Fixed Effects Yes Yes Yes Yes Yes Yes Yes Yes Observations 1118 1085 1085 1051 1051 1085 1059 795
- 45 -
Table 11b. IMF Program and Intensity of Sudden Stop
Robust z statistics are presented in parentheses. *** represents significance at 1 percent, ** at 5 percent, and * at 10 percent.
Dependent Variable Financial Account Outflow if a Sudden Stop Occurred (Tobit) GDP growth, lag -0.02*
[1.74] -0.02* [1.74]
-0.02* [1.71]
-0.02* [1.84]
-0.02* [1.80]
-0.02* [1.79]
-0.02 [1.61]
-0.01 [0.76]
Export growth, lag -5.3*** [2.92]
-5.52*** [2.98]
-5.53*** [2.98]
-6.12*** [3.48]
-6.12*** [3.48]
-5.51*** [2.97]
-5.57*** [3.00]
-5.69*** [2.92]
Debt servicing/ exports, lag
1.81*** [4.76]
1.95*** [5.00]
1.97*** [5.06]
1.22*** [3.07]
1.27*** [3.23]
1.93*** [4.95]
2.06*** [5.23]
2.68*** [6.02]
Domestic credit/ GDP, lag
0.03 [0.12]
0.03 [0.11]
0.02 [0.06]
0.08 [0.32]
0.10 [0.38]
0.03 [0.11]
0.08 [0.29]
0.17 [0.53]
External debt/ GDP, lag
0.10 [1.03]
0.11 [1.06]
0.11 [1.02]
0.11 [1.10]
0.11 [1.15]
0.10 [1.00]
0.19 [1.55]
0.10 [0.74]
IMF program existed in T, T-1
0.07 [0.73]
IMF program started in T, T-1
-0.17* [1.94]
-0.16* [1.73]
-0.19* [1.75]
-0.15 [1.45]
-0.15 [1.63]
-0.13 [1.39]
-0.24** [2.32]
Precautionary program in T,T-1
-0.22 [0.85]
Approved amount/GDP, lag
0.02 [0.76]
Approved amount/Debt, lag
0.003 [0.21]
IMF credit/debt, lag -0.96 [0.96]
IMF credit/GDP, lag -0.05** [2.34]
-0.04 [1.38]
Financial mobility lag -0.19*** [2.93]
ER regime: limited flexibility
-0.081 [0.38]
ER regime: managed float
-0.42** [2.14]
ER regime: free fall 0.28 [1.22]
Country Fixed Effects Yes Yes Yes Yes Yes Yes Yes Yes Year Fixed Effects Yes Yes Yes Yes Yes Yes Yes Yes Observations 1401 1361 1361 1328 1328 1361 1340 1087
- 46 -
Table 12. Dealing with the Endogeneity of Programs
Columns reporting probit results present marginal probabilities, based on the STATA command “dprobit”. Robust z statistics are presented in parentheses. *** represents significance at 1 percent, ** at 5 percent, and * at 10 percent. Dependent variable in first stage regressions (columns I and IV) is new IMF program in T or T-1; in second stage probit regressions (Column II and V) dependent variable is a dummy for first year of sudden stop; in Tobit regressions (Columns III and VI) the dependent variable is financial account outflow if a sudden stop occurred.
Program Determinants (Probit)
Sudden Stop Determinants (Probit)
Sudden Stop Intensity (Tobit)
Program Determinants (Probit)
Sudden Stop Determinants (Probit)
Sudden Stop Intensity (Tobit)
I II III IV V VI US aid, lag2 0.17***
[4.26] 0.17***
[4.20]
Voting pattern in UN, lag2
1.19 [1.03]
1.58 [1.33]
Reserve/import, lag2
0.001 [0.47]
0.002 [0.72]
Real growth, lag -0.01 [0.90]
-0.01*** [2.71]
-.03*** [2.64]
0.001 [0.13]
-0.003* [1.73]
-0.02* [1.88]
Real growth, lag2 -0.03*** [3.25]
-0.01** [2.50]
-0.03** [2.05]
Export growth, lag
2.28 [1.06]
-0.10 [0.27]
-2.12 [0.74]
2.26 [0.97]
-0.25 [0.59]
-1.29 [0.43]
Export growth, lag2
-1.25 [0.57]
-0.48 [1.39]
-4.34 [1.54]
Debt servicing/export, lag
1.04*** [3.03]
0.30*** [3.59]
2.81*** [5.78]
0.27 [0.73]
0.01 [0.10]
2.21*** [4.20]
Debt servicing/exports, lag2
1.46*** [3.23]
0.46*** [3.95]
0.89 [1.61]
Domestic credit/GDP, lag
0.31 [1.11]
0.21*** [3.52]
0.20 [0.60]
-0.18 [0.33]
0.10 [1.05]
-0.37 [0.61]
Domestic credit/GDP, lag2
0.52 [1.00]
0.11 [1.15]
0.61 [1.03]
External debt/GDP, lag
0.53*** [3.79]
0.08** [2.31]
0.41** [2.05]
0.46** [2.40]
0.06 [1.58]
0.28 [1.10]
External debt/GDP, lag2
0.07 [0.39]
0.01 [0.47]
0.13 [0.59]
probability of new program in T or T-1
-0.38*** [2.90]
-1.52** [2.27]
-0.37*** [3.06]
-1.30** [1.98]
Country Fixed Effects
Yes Yes Yes Yes Yes Yes
Year Fixed Effects
Yes Yes Yes Yes Yes Yes
Observations 1403 857 1098 1384 831 1084
- 47 -
Table 13. Probability of Sudden Stops and IMF Programs: Are There Threshold Effect?
Results present marginal probabilities, based on the STATA command “dprobit”. *** represents significance at 1 percent, ** at 5 percent, and * at 10 percent. Robust z statistics are presented in parentheses. Probability of an IMF program estimated as in the first column of Table 12. Dependent variable is a dummy for first year of sudden stop. Medium levels of debt is defined as when debt as a percentage of GDP is between 40 and 70; and debt is considered high if it exceeds 70 percent. Medium debt servicing level is defined as when debt servicing as a percentage of exports is between 13 and 27 percent; it is considered high if it exceeds 27 percent.
Threshold Effects for External Debt
Threshold Effects for Debt Servicing
Threshold Effects for Both External Debt and Debt Servicing
GDP growth, lag -0.005** [2.48]
-0.005** [2.53]
-0.005** [2.56]
Export growth, Lag -0.081 [0.22]
-0.14 [0.37]
-0.15 [0.39]
Debt servicing/exports, lag 0.27*** [3.35]
0.32*** [3.45]
0.28*** [3.48]
External debt/GDP, lag 0.06 [1.57]
0.08** [2.24]
0.07** [2.21]
Domestic credit/GDP, lag 0.20*** [3.41]
0.20*** [3.45]
0.20*** [3.47]
Predicted probability of a new IMF program in T or T-1
-0.45*** [3.21]
-0.34** [2.45]
-0.43*** [2.96]
Medium debt/debt servicing* probability of a program
0.09 [1.49]
0.003 [0.05]
0.08 [1.04]
High debt/debt servicing* probability of a program
0.12 [1.62]
-0.03 [0.43]
0.07 [0.75]
Country Fixed Effects Yes Yes Yes Year Fixed Effects Yes Yes Yes Observations 856 856 856
- 48 -
Table 14. Growth Regressions: Effects of Sudden Stops and IMF Programs
Full Sample Emerging Markets and 1990s
Dependent Variable Real Growth
Real Growth
Real Growth
Real Growth
I II III IV Per capita income, lag -0.3***
[2.69] -2.6*** [4.76]
-1.9*** [4.55]
-3.41*** [3.30]
Real growth, lag 0.24*** [5.93]
0.16* [3.66]
0.38*** [4.67]
0.13 [0.69]
Predicted prob of a Program in T, T-1 -1.21* [1.89]
0.38 [0.24]
0.48 [0.21]
1.59 [0.44]
Sudden stop _all years -1.49 [1.42]
-0.97 [0.97]
Sudden stop*Prob of Program -0.28 [0.14]
-2.37 [1.22]
Country FE no yes no no Time FE yes yes yes yes Observations 1206 1206 1209 218 R-squared 0.14 0.13 Probability of an IMF program estimated as in the first column of Table 12. *** represent significance at 1 percent, ** at 5 percent, and * at 10 percent. Robust t statistics are presented in parentheses. Columns (III) and (IV) were estimated to allow for the possibility that determinants vary when there is a sudden stop. The two-step “heckman” sample selection command in STATA was used. A first-stage selection equation estimated the probability of a sudden stop (using the specification in Table 12, column 1) and the estimated inverse-Mills ratio was added to the determinants of growth.
- 49 -
References Barro, Robert and Jong-Wha Lee (2004), “IMF Programs: Who is Chosen and What are the
Effects?” unpublished manuscript, Harvard University and Seoul National University. Berg, Andrew and Catherine Patillo (1999) “Are Currency Crises Predicable? A Test,” IMF Staff
Papers, Vol. 46, pp. 107-38 Broda, Christian. and Eduardo Levy-Yeyati (2003), “Dollarization and the Lender of Last
Resort,” in Eduardo Levy-Yeyati and Federico Sturzenegger (eds), Dollarization: Debates and Policy Alternatives, Cambridge, Mass.: MIT Press.
Calvo, Guillermo A. (2002), “Globalization Hazard and Delayed Reform in Emerging Markets,”
Economia (Spring), pp.1-29. Calvo, Guillermo A. (2005), “Crises in Emerging Market Economies: A Global Perspective,”
NBER Working Paper no. 11305 (April). Calvo, Guillermo A., Alejandro Izquierdo, and Luis F. Mejia (2004), “On the Empirics of
Sudden Stops: The Relevance of Balance-Sheet Effects,” NBER Working Paper no. 10520 (May).
Calvo, Guillermo A. and Carmen Reinhart (2000), “When Capital Flows Come to a Sudden
Stop: Consequences and Policy,” in Peter Kenen and Alexander Swoboda (eds), Reforming the International Monetary and Financial System, Washington, D.C.: IMF, pp.175-201.
Celasun, Oya and Rodney Ramcharan (2005), “The Design of IMF-Supported Programs: Does
the Balance Sheet of the IMF Matter?” unpublished manuscript, IMF. Chami, Ralph, Sunil Sharma and Ilhyock Shim (2004), “A Model of the IMF as a Coinsurance
Arrangement,” IMF Working Paper no. 04/219 (November). Cohen, Daniel and Richard Portes (2004), “Towards a Lender of First Resort,” CEPR Discussion
Paper no. 4615 (September). Cordella, Tito and Eduardo Levy Yeyati (2004), “Country Insurance,” Staff Papers 52 (special
issue), pp.85-106. Cordella, Tito and Eduardo Levy Yeyati (2005), “A (New) Country Insurance Facility,” IMF
Working Paper no. 05/23 (January). Eichengreen, Barry J., Kletzer, Kenneth M. and Mody, Ashoka (2005), “The IMF in a World of
Private Capital Markets” (March 2005). NBER Working Paper no. 11198. http://ssrn.com/abstract=684724. Forthcoming Journal of Banking and Finance.
Edwards, Sebastian (2005), “Capital Controls, Sudden Stops and Current Account Reversals,”
NBER Working Paper no. 11170 (March).
- 50 -
Dervis, Kemal and Ceren Ozer (2005), A Better Globalization: Legitimacy, Government and
Reform, Washington, D.C.: Center for Global Development.
- 51 -
Frankel, Jeffrey A., and Eduardo Cavallo (2004), “Does Openness to Trade Make Countries
More Vulnerable to Sudden Stops, or Less? Using Gravity to Establish Causality,” KSG Working Paper Number No. RWP04-038.
Frankel, Jeffery and Andrew Rose (1996), “Currency Crashes in Emerging Markets: An
Empirical Treatment,” Journal of International Economics, Vol. 41, pp. 351-66. Frankel, Jeffrey A., David Parsley and Shang-Jin Wei (2004), “Slow Passthrough Around the
World: A New Import for Developing Countries?” NBER Working Paper No. 11199 (Cambridge, Massachusetts: National Bureau of Economic Research).
Ghosh, Atish, Timothy Lane, Marianne Schulze-Ghattas, Ales Bulir, Javier Hamann, and
Alex Mourmouras (2002), “IMF-Supported Programs in Capital Account Crises,” Occasional Paper no.210, Washington, D.C.: IMF (February).
Gupta, Poonam, Deepak Mishra and Ratna Sahay (2003), “Output Response to Currency
Crises,” IMF Working Paper no.03/230. Independent Evaluation Office (2003), “The IMF and Recent Capital Account Crises,”
Washington, D.C.: IMF (October). Kaminksy, Graciela and Carmen Reinhart (1999), “The Twin Crises: The Causes of Banking
and Balance of Payments Problems,” American Economic Review 93, pp.473-500. Lane, Timothy and Steve Philips (2000), “Does IMF Financing Result in Moral Hazard?”
IMF Working Paper no. 00/168. Milesi-Ferretti, Gian Maria and Assaf Razin (1998), “Current Account Reversals and
Currency Crises: Empirical Regularities,” IMF Working Paper, WP/98/89. Mody, Ashoka, and Antu Murshid (2005), “Growing Up with Capital Flows,” Journal of
International Economics 65, pp.249-266. Mody, Ashoka, and Diego Saravia (2003), “Catalyzing Capital Flows: Do IMF Programs
Work as Commitment Devices?” IMF Working Paper 03/10 (Washington: International Monetary Fund). Forthcoming in Economic Journal.
Mody, Ashoka and Randall Stone (2005), “The Scope of IMF Conditionality: How Autonomous is the Fund?” unpublished manuscript, IMF
Husain, Aasim, Ashoka Mody and Kenneth Rogoff (2005), “Exchange Rate Regime
Durability and Performance in Advanced and Developing Countries,” Journal of Monetary Economics 52, pp.35-54.
Marchesi, Silvia (2001), “Adoption of an IMF Program and Debt Rescheduling: An
Empirical Analysis,” Centro Studi Luca D’ Agliano, Development Studies Working Paper no. 152.
- 52 -
Ostry, Jonathan D. and Jeromin Zettelmeyer (2005), “Strengthening IMF Crisis Prevention,”
IMF Working Paper WP 05/206 (November). Reinhart, Carmen and Kenneth Rogoff (2004), “The Modern History of Exchange Rate
Arrangements: A Reinterpretation,” Quarterly Journal of Economics 119, pp.1-48.
Truman, Edwin (2005), “International Monetary Fund Reform: An Overview of the Issues,” unpublished manuscript, Institute for International Economics (November).
- 53 -
Appendix A: Countries in the Sample and Sudden Stop Dates
Developing Countries: Sudden Stops (First Year) Emerging Markets: Sudden Stops (First Year)
Albania 1986, 1991 Argentina 1989, 2001 Armenia Brazil 1983, 2002 Bangladesh 1988 Chile 1983 Benin 1988 China Bolivia 1983 Colombia 1991 Botswana 1987, 1998 Czech Republic 1997 Bulgaria 1985, 1989 Egypt, Arab Rep. 1990, 2002 Burkina Faso 1989 Hong Kong, China Cambodia Hungary 1990, 1996 Cameroon 1990 India Central African Republic Indonesia 1997 Costa Rica 1982 Jordan 1993 Ecuador 1983, 2000 Korea 1986, 1997 El Salvador Malaysia 1998 Estonia Mexico 1983, 1987, 1995 Ethiopia 1994, 1996, 1999,
2001 Morocco 1995, 2003
Ghana 2002 Pakistan 1998 Guatemala 1984 Peru 1983, 1986 Guinea 1989, 1997, 2002 Philippines 1998, 2002 Haiti 1993, 2003 Poland 1982, 1988, 1990, 1994 Honduras 1989 Singapore Jamaica 1986 South Africa 1985 Kenya 1992, 2002 Thailand 1997 Lao PDR 1988, 1998 Turkey 1994, 1998, 2001 Latvia Lesotho Lithuania Madagascar 1992 Malawi 1981 Mali 1987, 1992 Mauritania 1993, 1996 Mauritius 1985, 2001 Mongolia 1991 Mozambique 1988, 2002 Myanmar Namibia 2000 Nepal 2001 Nicaragua 1991 Niger 1984, 1991, 1993 Panama Paraguay 1988 Romania 1981, 1988 Rwanda 1988, 1994, 2001 Senegal 1996
- 54 -
Appendix A (Continued): Countries in the Sample and Sudden Stop Dates
Developing Countries: Sudden Stops (First Year)
Emerging Markets: Sudden Stops (First Year)
Sierra Leone 1986 Sri Lanka Sudan 1985 Swaziland 1987 Taiwan POC Tanzania 1994, 2001 Togo 1992 Trinidad and Tobago 1986, 1988, 1995 Tunisia Uganda Ukraine Uruguay 2002 Vietnam Zambia
- 55 -
Data Appendix Sudden stop Identified as described in the text using the annual data on Financial
account balance from the BOP database, IMF Real growth Calculated using data from the WEO, IMF Export growth Calculated using data from the WEO, IMF Domestic credit IMF's IFS database External debt World Bank's GDF database Debt servicing World Bank's GDF database IMF program Data from the Policy Development and Review Department of the IMF Precautionary pProgram Data from the Policy Development and Review Department of the IMF IMF Credit IMF's BOP database Financial Account Balance IMF's BOP database FDI IMF's BOP database Portfolio flows IMF's BOP database Other capital flows IMF's BOP database Current account balance IMF's BOP database Capital formation IMF's IFS database Financial integration From Mody and Murshid (2005), updated using data from the IMF Exchange rate regime From Reinhart and Rogoff (2004) Currency crisis From Kaminsky and Reinhart (1999), Berg and Patillo (1999), Frankel
and Rose (1996), Milesi-Ferreti and Razin (1998) and Frankel and Wei (2004)
Banking crisis From Demirguc-Kunt and Detriagiahe supplemented with Goldstein, Kaminsky and Reinhart (2000), and Caprio and Klingebiel (1996)
US aid OECD's DCA database U.N. voting pattern From the website http://home.gwu.edu/~voeten/UNVoting.htm Emerging markets J.P. Morgan's MSCI US growth rate IMF's IFS database US treasury bill rate IMF's IFS database