I. Introduction and Summary

The economic crisis that hit Russia and dried up the financing to most developing countries for the remainder of 1998, is one in a series of major economic shocks that have affected emerging economies over the last five years, including the “tequila” crisis of late 1994, and the Asia crisis that began in 1997. Many observers have attributed the rapid spread of these crises to contagion. In line with this conclusion the IMF instituted, in April 1999, a new mechanism, the contingent credit line or CCL, to provide countries with additional liquidity to overcome the impact of contagion.

Opponents of the extension of such international liquidity tend to argue that this leads to moral hazard: excessive risk taking by (potential) recipients of such insurance. It is thus of particular interest to find out to what extent the depth and incidence of crises reflect solvency related fundamentals, and whether and to which extent they reflect a lack of liquidity. The former should in principle be dealt with through underlying policy measures that reestablish solvency, the latter, if present, either through the extension of liquidity by an international body or lender of last resort (Fischer 1999) or the buildup of additional reserves (Feldstein 1999).

Sachs, Tornell and Velasco (1996), hereafter STV, provides a useful methodology to examine vulnerability during general crisis periods, and in which risks of contagion are highest. The STV methodology focuses on external vulnerability during a window of about five months following the outbreak of a major crisis (the Mexican crisis in their application), i.e. the period during which countries are most prone to contagion, and most vulnerable to the lack of international liquidity that frequently follows eruption of major crises. This approach is consistent with stylized facts regarding the data on emerging market bond spreads, which tend to peak following a major crisis in a leading emerging market economy, and frequently coincide with a dearth of international financing.

To establish credibly which key variables are relevant, we first examine which sets of variables are able to predict the 1998 crisis out-of-sample. This is especially important given the relatively small number of observations implied by the STV methodology, and the poor results of various methods in out-of-sample tests for the 1997 crisis (Berg and Pattillo, 1998).² The key sets of variables we use are those used by Tornell (1998) (referred to as STV/T) in his extension of STV to the 1997 crises (lending boom, real exchange rate appreciation, reserve adequacy, and “strong fundamentals”), and the variables that feature in an Early Warning System model developed by Fund staff (EWS hereafter) (current account deficit, short term debt over reserves, export growth, real exchange rate appreciation, and the change in reserves) as presented to the IMF Board.³

The out-of-sample projections for 1998 yield very clear results. Making only relatively minor a priori data improvements (regarding the definition of data such as reserves and real exchange rates) the results indicate that the combined STV/T set of variables, as estimated for 1994 and 1997 crises, functions as a contraindicator for the 1998 crisis. When the key variable, the lending boom, is adjusted for an obvious flaw (namely that it exaggerates vulnerability of those countries that already underwent a banking crisis and cleaned up their banking systems), it still does not perform well. This contrasts with the results obtained when using the EWS variables. Predictions of the rankings of the vulnerable countries based on the EWS variables match more closely the actual ranking, although the extent of vulnerability of a number of countries is overestimated.

Estimations of the crisis periods, combined and separate, confirm that the three crisis periods can be largely explained by a parsimonious equation of three of the five EWS variables. Two of the three variables are fundamental variables that feature very prominently in evaluations of countries’ solvency prospects (the overvaluation of the exchange rate and current account deficit). Of these two, the first is generally found to be significant, whereas the second usually is not (Kaminsky, Reinhart and Lizondo, 1998 or KLR). The third (short-term debt over reserves) is the liquidity-related variable that has been emphasized as the most prominent indicator of liquidity conditions (Greenspan 1999, Sachs and Radelet 1998, and recent internal guidance for IMF staff).

Having obtained such a parsimonious and fairly stable model allows us to conduct a number of additional tests. In particular, we investigate the role of domestic credit related variables. We show that, while the 1994 and, to some extent, 1997 crises were driven by credit to the banking sector, credit to the government sector was a key explanatory variable in the 1998 crisis. In addition we conduct a number of tests on liquidity-oriented variables to examine whether alternative specifications, such as various ratios of reserves over money, and the traditional reserve coverage of imports perform better—which they clearly do not. We furthermore test for alternative specifications for the current account which indicate that financing deficits through FDI might limit vulnerability, but the effect is very weak. Finally we test for the relevance of having a Fund program. None of the studies in the literature as reviewed by KLR report testing of such an effect, even though having Fund program might be expected to have a mitigating impact on crises. We find a very significant impact of having a Fund program on reducing the depth of a crisis. Having a program reduces the crisis indicator by 10 percentage points. This compares to an average level of the crisis indicator of 5.2 in the tequila crisis, 18.8 in the Asia crisis and 8.9 in the 1998 crisis.⁴ This result is obtained despite the fact that two countries deeply affected by the 1998 crisis (Russia and Zimbabwe) had Fund programs. Moreover, the results pass various robustness tests.

This suggests that Fund programs reduce vulnerability even beyond the positive effect they may have on the fundamentals included in the specification. This result could indicate that fundamentals not included in the estimations are sounder in countries with Fund programs (e.g. confidence in countries taking appropriate adjustment measures, solid structural policies). It could also indicate that having a Fund program creates the expectation of additional cushion of liquidity or has a reputational or catalyst effect.

The estimation results suggest that higher liquidity (as presented by the level of reserves over short-term debt) can offset weak fundamentals (as represented by the current account deficit and the appreciation of the exchange rate) and limit the vulnerability of countries in periods of contagion.⁵ The existence of such an ability to offset weak fundamentals by high liquidity contrasts sharply with a fundamentals only view of the world, in which solvent countries are always able to exploit their solvency by borrowing their way out of a potential crisis, and thus have no real need for a cushion of liquidity. Apparently, “solvency is very much like honesty: it can never be fully certified, and proofs are slow to materialize,” to quote Calvo (1996), and liquidity is required while investors learn whether the crisis effected countries are solvent or not.

The relation between reserves and fundamentals we find, can be used to establish some rules of thumb or guidance for judging reserve adequacy. Making use of the fact that the crisis index in STV is continuous, and that our empirical results are quite parsimonious, we calculate the reserve coverage of short-term debt that is compatible with reducing external vulnerability. The results imply a differentiated reserve coverage depending on easily observable variables (the real exchange rate appreciation over the last 48 months and the current account deficit), and having a Fund program, which is presented in a straightforward set of tables (Tables 13a-c). The results indicate as a rule of thumb, that for countries with real exchange rates that are not significantly misaligned, and modest current account deficits, a reserve coverage of total short-term debt of one (as suggested by Greenspan 1999), is broadly consistent with avoiding sizable contagion, especially for countries with Fund programs.

It is in this regard worthwhile to emphasize a simple but important logical implication: adequate reserve coverage of short-term debt can be accomplished either by increasing reserve levels or by reducing exposure to short-term debt. Clearly, countries need not deter short-term debt. Rather than attempting to control short-term capital inflows, countries can pursue increased reserves.

Targeting reserve cover of short-term debt has a number of other implications as well. First it provides an incentive for central banks to avoid explicit or implicit biases in favor of short-term inflows (as in Korea and Thailand prior to their crises), as such inflows would require higher reserve levels. Second, if reserve management focuses on coverage of short-term debt by remaining maturity as a key benchmark, this builds in incentives to guide public debt management in such a way as to lengthen the maturity and to avoid bunched repayments. Thus focussing on such a criteria may reduce the need for debt management guidelines (Greenspan 1999). Moreover, focus on debt guidelines this should not distract from improving data—comprehensive data on derivative positions of central banks and short-term debt, while part of newly agreed data dissemination standards, are still not yet widely available, and the reliance of this paper on creditor data on debt testifies to that.⁶

The paper is organized as follows. Section II presents the out-of-sample prediction of the STV model for 1998 based on the parameters obtained in 1994 and 1997. Section III tests the EWS variables. Section IV reports on detailed tests on alternative liquidity ratios and additional and alternative variables. Section V discusses implications for reserve management, given the level of short-term debt and other macroeconomic variables, while Section VI characterizes the three crisis periods based on the estimation results. The paper concludes with health warnings.

II. Replicating and Testing STV/Tornell

III. EWS Variables

IV. Fund Programs, Other Variables and Robustness

Having found a parsimonious result that satisfactorily predicts the 1998 crisis out-of-sample (and out-of-sight in the sense that it is based on variables that were selected before the onset of the Russia crisis), and which shows stable coefficients for the three sub periods, we can use this result to test alternative explanations, without being overly exposed to the dangers of finding spuriously correlated results. It is in particular of interest to test the following: the role of Fund programs in mitigating or exacerbating crises; alternative specifications for the liquidity ratios—this has potentially important implications for reserve targets; the role of credit expansion to the public sector—to capture the role of fiscal expansion; alternative specifications for the current account deficit which focus on exports as a denominator rather than GDP and the vulnerability mitigating role of FDI in financing such deficits. Moreover, we test the robustness of the results by excluding vulnerable countries, and test for fixed and regional effects.

C. Credit Variables

While Sections II and III established that private sector credit growth did not perform well in predicting the 1998 crisis (even when the variable is respecified to encompass possible effects of banking sector reforms), this does not completely rule out the relevance of credit in contributing to external vulnerability. In addition we have not explained the significance of the lending boom variable in the first crisis period and the switch in the signs between crisis periods from which even the corrected lending boom variable suffers. To tackle these issues we first tried to correct the scaling. Second, we examined the impact of credit to the public sector, especially since the 1998 crisis was centered in countries with large domestically financed fiscal deficits.

LBC scaled by reserves

In the STV/T approach, the lending boom is measured as the percent change of the credit to the private sector (in real terms) over a four year period, regardless of the relative importance of private sector credit to the economy. As an alternative, we used the increase in credits to the private sector in real terms, scaled by the level of reserves, for which data are available monthly, and which are indicative of the resources available to the monetary authorities to prevent a banking sector crisis to spillover into an external crisis:

$\mathit{\text{LBRPC}}=\frac{\frac{L_i}{{\mathit{\text{CPI}}}_i}-\frac{L_{i-48}}{{\mathit{\text{CPI}}}_{i-48}}}{e_i.\text{Re}{s}_i}$

where i is a given month (October 1994, June 1997, and June 1998 in our estimation); L represents credit to the private sector, CPI is the normalized consumer price index, e the exchange rate, and Res the level of international reserves. In line with the adjustment of LB to LBC we use, instead of the increase in credit over 48 months, only the increase in credit since the previous crisis for those countries that underwent a banking sector reform. Table 10 presents results obtained with this variable combined with the three core variables. LBRPC clearly performs better than LBC, when combined with the EWS variables, but still is not significant over the three periods combined. Noteworthy is that it is significant with the expected positive sign for the first two periods, but just as LBC its sign in the Russia crisis period is negative.

Credit to the government

To test the role of credit to the government we constructed another variable, along the lines of LBRPC for the banking sector credit to the government:¹³

$\mathit{\text{LBGOVP}}=\frac{\frac{{\mathit{\text{LGOV}}}_i}{{\mathit{\text{CPI}}}_i}-\frac{{\mathit{\text{LGOV}}}_{i-48}}{{\mathit{\text{CPI}}}_{i-48}}}{e_i.\text{Re}{s}_i}$

where CPI, e and Res are the same as in the previous equation, and LGOV represents credits to the government sector. It is noteworthy that when tested for the 1998 crisis alone, this variable is very significant, has the correct sign and leads to a jump in the R-squared that indicates that about 90 percent of the variation in the crisis index is explained. In 1998, crisis countries had a very high LBGOVP variable, as summarized in Table 3, which ranks the countries with the highest LBGOVP variable.

Table 1.

The Five Most and Least Vulnerable Countries in 1998

(Comparison EWS Forecasts and Actual Values)

Source: Estimation based on Equation (2), Table 6.

Table 1.

The Five Most and Least Vulnerable Countries in 1998

(Comparison EWS Forecasts and Actual Values)

	Country	Fitted Crisis Index With EWS	Actual Crisis Index in 1998	Actual Rank
1	Zimbabwe	68.9	82.2	2
2	Russia	53.7	87.2	1
3	Pakistan	35.4	0.1	11
4	Argentina	29.7	-0.9	16
5	Brazil	26.4	32.4	3
18	Korea	3.1	-3.4	18
19	Malaysia	1.5	-9.8	21
20	Sri Lanka	-0.6	1.5	9
21	Jordan	-4.7	0.2	10
22	India	-6.5	-2.1	17

Source: Estimation based on Equation (2), Table 6.

View Table

Table 1.

The Five Most and Least Vulnerable Countries in 1998

(Comparison EWS Forecasts and Actual Values)

	Country	Fitted Crisis Index With EWS	Actual Crisis Index in 1998	Actual Rank
1	Zimbabwe	68.9	82.2	2
2	Russia	53.7	87.2	1
3	Pakistan	35.4	0.1	11
4	Argentina	29.7	-0.9	16
5	Brazil	26.4	32.4	3
18	Korea	3.1	-3.4	18
19	Malaysia	1.5	-9.8	21
20	Sri Lanka	-0.6	1.5	9
21	Jordan	-4.7	0.2	10
22	India	-6.5	-2.1	17

Source: Estimation based on Equation (2), Table 6.

Table 2.

Sample Averages for Explanatory Variables

Table 2.

Sample Averages for Explanatory Variables

	CrisindGld	RERINS	CA/GDP	STD/R
With Fund program	7.93	-2.94	3.41	96.78
Sample	11.04	-9.01	3.07	82.01
No Fund program	13.11	-13.07	2.84	72.16

View Table

Table 2.

Sample Averages for Explanatory Variables

	CrisindGld	RERINS	CA/GDP	STD/R
With Fund program	7.93	-2.94	3.41	96.78
Sample	11.04	-9.01	3.07	82.01
No Fund program	13.11	-13.07	2.84	72.16

Table 3.

Five Highest Increases in Credit to the Government (in percent)

Table 3.

Five Highest Increases in Credit to the Government (in percent)

	LBGOVP
Russia	182
Zimbabwe	149
Brazil	52
Pakistan	49
Indonesia	37

View Table

Table 3.

Five Highest Increases in Credit to the Government (in percent)

	LBGOVP
Russia	182
Zimbabwe	149
Brazil	52
Pakistan	49
Indonesia	37

However, in the two earlier periods, credit extension to the government is not only insignificant but also has the wrong sign. I.e. the variable performs opposite to LBRPC. In 1994 and 1997 most crisis countries had a negative LBGOVP variable (like Brazil and Mexico in 1994, Indonesia, Korea, Malaysia and Thailand in 1997). In 1994, credit to the government, decreased rapidly in Mexico and Brazil, with the government switching to non-bank financing thus freeing bank credit for private lending (see also Dornbusch 1997). Such phenomena thus seem to have contributed to the high and significant LBRP for 1994, while reverse pressures were at play in 1998, with governments to some extent at least crowding out private lending.

This would suggest that total credit would be a better indicator but this is not so.¹⁴ Thus the data suggest that rapid credit expansion to either the private or government sector can significantly add to vulnerability, while a contraction of credit in the other sector does not provide relieve.

D. Alternative Specifications of the Current Account Variable

Scaling with exports

In cross country analysis and as a shorthand indicator for external vulnerability, the current account balance is usually scaled by GDP to reflect its relation to the overall size of the economy. This practice does not take into account the degree of openness, which is essential to evaluate the capacity of a country to repay its debt. Therefore we also tested the current account as a percentage of exports of goods and services. However, when tested, the current account deficit scaled by GDP was more significant by a wide margin (Table 9).

The role of FDI in current deficits

FDI has become an increasingly important source of funding for current deficits in emerging economies. For example, Brazil’s current deficit is on balance largely financed by FDI, and the same applies for a number of other Latin American countries (see Table 4). It has been frequently argued that such FDI mitigates vulnerability, as the stock can not leave the country quickly, profits remittances could be countercyclical, FDI contributes to higher (export) growth, and a drop in non-portfolio FDI often leads to a significant decline in imports of investment goods. Therefore we also tested for the significance of FDI over GDP. This variable was not significant at the 10 percent level, although it had the correct sign, suggesting only a limited reduction in vulnerability as a result of FDI financing of the deficit. It should of course be noted that countries with significant FDI tend to have less short-term debt, and thus some of the positive effects of FDI are indirectly captured in that variable.

Table 4:

Current Account Deficit over GDP Before and After FDI for Selected Countries with Substantial FDI

Table 4:

Current Account Deficit over GDP Before and After FDI for Selected Countries with Substantial FDI

	CA/GDP	(CA-FDI)/GDP
Colombia 1998	7.24	-0.38
Peru 1998	5.22	0.58
Chile 1998	5.26	-1.76
Brazil 1998	4.21	1.76
Argentina 1998	2.91	0.86
Venezuela 1998	-5.30	-11.05

View Table

Table 4:

Current Account Deficit over GDP Before and After FDI for Selected Countries with Substantial FDI

	CA/GDP	(CA-FDI)/GDP
Colombia 1998	7.24	-0.38
Peru 1998	5.22	0.58
Chile 1998	5.26	-1.76
Brazil 1998	4.21	1.76
Argentina 1998	2.91	0.86
Venezuela 1998	-5.30	-11.05

E. Non-Linear Functional Forms

The core equation for the crisis index is linear in the dependent variables. However, the fact that two of the variables are ratios implies that the relation vis à vis the nominator in the ratios is non-linear (see Figure 1). To further examine any non-linearity we tested various alternative specifications, including logarithmic, exponential and multiplicative. Only the multiplicative specification in short-term debt over reserves performs well, notably the following (Table 11, eq. 6):

Trade-off between Liquidity and Fundamentals, Linear Function (crisisindex=0, RERINS=0)

Citation: IMF Working Papers 1999, 088; 10.5089/9781451851144.001.A001

• Download Figure
• Download figure as PowerPoint slide

View Full Size

Trade-off between Liquidity and Fundamentals, Linear Function (crisisindex=0, RERINS=0)

Citation: IMF Working Papers 1999, 088; 10.5089/9781451851144.001.A001

• Download Figure
• Download figure as PowerPoint slide

Trade-off between Liquidity and Fundamentals, Linear Function (crisisindex=0, RERINS=0)

Citation: IMF Working Papers 1999, 088; 10.5089/9781451851144.001.A001

• Download Figure
• Download figure as PowerPoint slide

$\text{Crisisind}={(-{0.45}^{*}\text{RERINS}+{1.52}^{*}\text{CA}/\text{GDP}+{0.13}^{*}\text{STD}/R-13.2\text{FUNDP})}^{*}(\text{STD}/R)/100-2.5$

This specification explains well over 80 percent of the variation, compared to 70 percent for the standard specification that is linear in the dependent variables (Table 9, Eq. 1). The significance of most terms increase. The interpretation of this result is discussed in Section V. The main difference with the linear specification is that a high ratio of debt over reserves (limited liquidity) increases the vulnerability for a deterioration in the fundamentals. I.e., weaknesses reinforce each other at the margin.¹⁵ When this specification is estimated for 1994 and 1997, it predicts the 1998 crisis out-of-sample better than the specification that is linear in the dependent variables, as measured by Spearman’s rank correlation coefficient, which becomes 0.76.

V. Policy Implications for Crisis Prevention

Greenspan (1999) suggests that it is necessary to pay closer attention to the ratio of short-term debt to reserves and suggest that a ratio of one may be an appropriate target. In the same vein, the IMF’s new CCL facility specifies that close attention needs to be paid to short-term debt by remaining maturity. Our parsimonious and rather robust result underscores the importance of this variable and lends itself well for evaluating the desirable level, as the equation can be solved for the ratio of reserves over short-term debt. The result implies that the higher the reserve coverage, the less vulnerable a country is. It also suggests that there is a trade-off: that high levels of reserves reduce vulnerability and can, to some extent, help overcome the vulnerability caused by high current account deficits and overvalued exchange rates. I.e. it suggest there is a choice between vulnerability and liquidity.

Tables 13a-b quantify the trade-off embodied in the core linear equation with the Fund program dummy (Table 9, equation 1) for ranges of current account deficits and real effective exchange rates. The results are illustrated in the Figure below. The Figure illustrates the extent liquidity can offset poor fundamentals based on the definition of debt used in the estimations (BIS data on claims of banks) and an inferred relation based on total short-term debt by remaining maturity—the size of debt reported by the BIS amounts in our sample to about seventy percent of the total short-term debt on a remaining maturity basis.¹⁷

The results show that if the real exchange rate is not overvalued and the current account has a modest surplus, a ratio of reserves to total short-term debt of 1 would result in a zero crisis index for countries without a Fund program. Countries with a Fund program or countries willing to tolerate a modest crisis index (a weighted reserve/exchange rate loss of 10 percent), the data imply that a modest current deficit (2-3 percent) and a slightly appreciated exchange rate (6-9 percent) could be maintained. The results also indicate that, especially for very unbalanced real exchange rates, and significant current deficits, little remedy in reserves is available, as the liquidity ratios implied are very sizable and presumably unsustainable from a fiscal perspective.

The non-linear specification has broadly similar implications. Avoiding a crisis altogether requires a somewhat higher ratio of reserves to short-term debt for countries without Fund program (Figure 2). For slightly higher levels of the crisisindex, the ratio is still about 1, as Table 13c illustrates—Table 13c quantifies the impact on the crisisindex of different ratios of reserves over debt with varying degrees of exchange rate appreciation, assuming a current account deficit of 3 percent.

Trade-off Fundamentals and Liquidity, Multiplicative Function (Crisind=0, REER=0)

Citation: IMF Working Papers 1999, 088; 10.5089/9781451851144.001.A001

• Download Figure
• Download figure as PowerPoint slide

View Full Size

Trade-off Fundamentals and Liquidity, Multiplicative Function (Crisind=0, REER=0)

Citation: IMF Working Papers 1999, 088; 10.5089/9781451851144.001.A001

• Download Figure
• Download figure as PowerPoint slide

Trade-off Fundamentals and Liquidity, Multiplicative Function (Crisind=0, REER=0)

Citation: IMF Working Papers 1999, 088; 10.5089/9781451851144.001.A001

• Download Figure
• Download figure as PowerPoint slide

Both results confirm that a ratio of reserves to short-term external debt of about one is a useful benchmarks. In translating these empirical results into predictions good note should be taken of the fact that both reserves and short-term debt used here are gross concept. As our adjustments for South Africa indicate, the neglect of non-reserve assets is likely to exaggerate vulnerability for countries with significant non reserve assets in relation to their debt. The use of BIS data by remaining maturity implies that we include items in short-term debt such as foreign currency deposits by residents with the branches of banks whose headquarters report their positions to the BIS. This may for example contribute for example to the overestimation of the vulnerability of Argentina (Table 1). Moreover some of the debt maybe owed by foreign invested enterprises reducing vulnerability. Finally reserves may be undermined by liabilities (both contingent and contingent ones) and derivative positions. Publicly available data do not at present allow for a more refined estimation of the effect of such positions.

VI. A Characterization of the Three Crises

The estimation results for the three periods allow for a broad characterization of the three crisis periods:

• All three crises were characterized by a strong liquidity-related element—and not just for the Asia crisis where the effect of this variable was emphasized. The ratio of short-term debt to reserves is the key liquidity indicator, and empirically outperforms the ratio of M2 to reserves and other money-based indicators and the traditional import based indicators with a wide margin.

• The real effective exchange rate played a key role in the 1998 crisis. Countries which experienced a crisis in 1998 (Russia and Brazil, Colombia to a lesser extent) had experienced a sharp appreciation of their real exchange rate index (respectively 35.7%, 37.5% and 24.5%). The Asia countries, which experienced a recovery during this period, underwent a depreciation of their real exchange rate, and this is associated with a negative crisis index for 1998.

• For the three crisis periods combined credit did not have a significant effect. However, for subperiods credit played a significant role. Credit to the government sectors aggravated the 1998 crisis. Estimates for this period, when including variables capturing such credit extension explain about 90 percent of the variation in the crisis index.

• Credit to the private sector contributed to the 1994 crisis. Not only as an indicator of a banking sector weakness, but also as an indicator of failed policies to maintain low banking sector interest rates by substituting government credit for non-bank financing (e.g. Dornbusch 1997) resulting in even higher private credit expansion.

• Credit to the private sector also played a role in the 1997 crisis. However, the extension of such credit as measured in this study (over the preceding 48 months) was closely correlated with the real exchange rate appreciation over the same period and the size of the current deficits, which contributed consistently to all crises. Moreover regional spillover effects were unusually strong, as indicated by the regional dummy for Asia.

VII. Concluding Remarks

As with investing in stock market funds, crisis models require a clear warning: “Past performance is no guarantee for future performance.” While the results we obtain are relatively robust and yield very interesting insights, especially about the desired level of reserves, we feel that two specific provisos need to be emphasized.

First, regarding liquidity. The results reflect the present state of play: this is subject to change. An expansion in the liquidity available to international arbitrageurs such as banks trading for their own portfolio and hedge funds, is likely to result in increasingly sharp reactions to deviations in fundamentals from sustainable levels. In contrast, the higher the liquidity of the IMF, and the more its lending is based on providing liquidity support to countries with sound fundamentals, the less likely are crises based in a lack of liquidity in countries (potentially) willing to embrace or seek Fund support.¹⁸ However, without fundamental changes in the system, a level of reserves equal to or higher than short-term debt by remaining maturity appears an advisable target for countries with broadly balanced exchange rates and modest current account deficits. Countries with higher deficits, even when financed through FDI, and appreciated exchange rates need significant higher reserve cover to limit the impact of contagion on their economies.

Second, solvency is a very complex function of economic and political variables. The types of models we are estimating capture only a few of the relevant variables. Thus in the 1994 and 1997 crises government credit did not rear its head, but in 1998 it was important. Sound predictions and evaluations require in-depth solvency and liquidity evaluations. Simplified models, as this contribution, are no substitute for discriminating and detailed analysis of the vulnerability of countries. These models are aids in the identification of the more relevant variables and critical magnitudes, but other latent variables maybe essential to future crises.