Updating a Statistic on Leaders’ Loss of Office

First we need to define a currency crash. Cooper counted anything more than 10 percent as a devaluation episode. But the world changed in the 1970s and 1980s, and depreciations of that magnitude have become commonplace. For a high-inflation country, one would not want to say that a new currency crisis occurs every month. So I use the following definition:²

• The devaluation must be at least 25 percent, on a cumulative 12-month basis.

• The devaluation must represent an acceleration of at least 10 percentage points, relative to the rate of depreciation in the 12 months before that.

• It must have been at least three years since the last currency crisis.

By these criteria, looking at a sample of 103 developing countries over the period 1971–2003, we found 188 currency crashes. In these countries, the person holding the position of chief executive changed within 12 months of the devaluation 27 percent of the time. The standard of comparison that we use normally is all other 12-month periods: the leader changed 21 percent of the time normally. Thus, devaluation increases by an estimated 32 percent the probability of the executive losing his or her job. The difference is statistically significant only at the 13 percent level.³ However, it may be that countries that tend generally to instability are overrepresented in the crisis group, so that political turnover is more common in this group and is not necessarily the result of currency crises. If we narrow the standard of non-devaluation comparison periods to the set of countries that have experienced a currency crash at some point during the sample period, on the grounds that these are more comparable to the crisis episodes, we find that the increase in job loss among devaluers now becomes almost statistically significant at the 1 percent level.⁴

We then narrowed the window to a half year. Now the chief executive lost office 19.1 percent of the time, as opposed to 11.6 percent of the time otherwise. In other words, a currency crash increases the probability of a change in the top leadership within the following six months by 1.7 times. This time the difference is statistically significant not only at the 10 percent or 1 percent levels, but at the 0.5 percent level as well, regardless of whether the entire set of countries is used as the standard of comparison.

We also looked at whether the finance minister or central bank (CB) governor—whoever held the office of the country’s governor of the IMF—lost his or her job. Here, even using the longer 12-month window (and even with only five years of data: 1995–99), the effect is statistically significant. In the year following a currency crash, the holder of this position changed 58.3 percent of the time. In other years during this period the rate of turnover was 35.8 percent. By this measure the finance minister or CB governor was 63 percent more likely to lose office. The difference is highly significant statistically (at the 0.001 level).⁵

When we segregate countries according to three income levels, we find that the phenomenon is chiefly one of middle-income countries. Within the class of poor countries, the increase in turnover of the leader is not statistically significant, and among rich countries there were no cases of a leader losing office within a year of a devaluation.

We also tried segregating countries according to three kinds of political structure: presidential democracy, parliamentary democracy, and nondemocracy. Our expectation was that we would find that the effect of devaluation on leadership turnover would be greater among parliamentary democracies than among presidential democracies. The logic was that in any given year the latter might not have a scheduled election, or, if they did, a term limitation might prohibit the incumbent from running for reelection. We found, instead, that the job-loss rate was much higher and more significant in the case of devaluations occurring in presidential democracies.⁶

We examined whether a loss of reserves equally as large results in job loss as often as a devaluation. Such episodes also tend to be unpleasant (often implying monetary contraction and recession; for example, Argentina in 1995 and 1999). Apparently they do not carry the same political costs: however their effect on leaders’ job loss was not significant.

What is it about devaluation that carries such big political costs? How is it that a strong ruler like Indonesia’s Suharto can easily weather 32 years of political, military, ethnic, and environmental challenges, only to succumb to a currency crisis?

Possible Sources of Political Costs of Devaluation

Currency crises are often accompanied by sharp recessions. Thus, an obvious interpretation, which we will consider further, is that leaders are punished by their constituents when the performance of the economy is poor. But before proceeding on the assumption that the loss of ministerial jobs is a reflection of unemployment and depressed activity throughout the economy, let us consider the possibility that the costs of a devaluation may be more political than economic. First, there is the possibility that elections cause currency crashes, rather than the other way around. Second, it could be that IMF programs or other austerity programs are unpopular in general and that the devaluations are an incidental aspect of this. Third, it could be that the leaders in question had made public promises in advance not to devalue and that they were punished for breaking these promises, regardless of subsequent economic performance.

What do I mean by the first possibility—that elections cause devaluations, rather than the other way around? It is striking in how many of the major crises of the 1990s, even though trouble began during the run-up to a major regularly scheduled national election, the worst speculative attack and currency crash came soon after the election. This describes Mexico in 1994, Korea in 1997, and Brazil in 1998–99. In an earlier era, one would have guessed that election-motivated macroeconomic expansion—the famous political business cycle—explained the need for a subsequent devaluation. But that explanation does not fit the experience of the 1990s. Macroeconomic expansion in these election campaigns was limited.⁷

A better explanation is that devaluation is politically costly to leaders, and so in an election year they try to postpone it—hoping to get reelected, that the crash happens on their successors’ watch rather than theirs, or that something will turn up in the meantime to improve the balance of payments.⁸ A related hypothesis is that because a devaluation uses up scarce political capital, it is more likely to be undertaken by a new leader with a strong mandate, especially in a visible crisis, and especially if he can blame it on his predecessor. Edwards (1994, Table 5) reports that devaluations occur disproportionately often during the first two years after a transfer of government: 77.3 percent of devaluations among presidential democracies (that is, those with prescheduled elections) and 70.0 percent among parliamentary democracies. This is a topic worth exploring, but not here: my calculations about the frequency with which ministers lose their jobs in the year after a devaluation were careful to start the clock the day after the devaluation, so that cases in which the devaluation came soon after an electoral change are not included in the statistics.

The second possibility I mentioned is that devaluations are a proxy for unpopular IMF austerity programs or other broad reform packages. IMF-associated austerity programs have often resulted in popular unrest. For example, riots following food-subsidy cutbacks contributed to the overthrow of President Nimeiri of Sudan in 1985.⁹

One can test the proposition that devaluations are acting as a proxy for unpopular IMF austerity programs by conditioning our previous calculation on the adoption of IMF programs. We created a dummy variable to represent cases where an IMF program was initiated within three months on either side of the devaluation.¹⁰ The IMF program variable does not seem to raise the frequency of leader job loss relative to devaluations that did not involve an IMF program. Thus, it is not surprising that conditioning on the IMF dummy variable has no discernible effect on the frequency of leader turnover: 21.05 percent of the time for the cases with an IMF program; 21.92 percent of the time for the ones without. In both cases, it is similar to the overall rate of job loss following devaluations (19.05 percent) in the complete sample and is still almost double the 11.6 percent rate in normal times.

That leaves the third noneconomic explanation, that the ministers in question have made public promises in advance not to devalue, and that they feel it necessary to resign or are punished for breaking these promises, regardless of subsequent economic performance. In many cases the commitment to the peg is explicitly reaffirmed by top policymakers and political leaders in the months immediately prior to the devaluation. Perhaps such ill-fated promises are made because the minister is duplicitous or at least is ignorant of the speculative pressures he or she is up against. More likely they are too attached to the peg psychologically to let go; many of the currency crashes of the 1990s occurred in countries where governments had a lot invested in the peg, because exchange-rate-based stabilizations earlier had been the successful and popular means of ending a 1980s cycle of high inflation, even hyperinflation.

Perhaps it is even better to regard the public commitments as sincere expressions of a strong desire to maintain the peg. The ministers may realize that events could force the abandonment of the exchange rate policy, if speculative pressures accelerate and it develops that reserves are about to run out, leaving little other option. And they may realize that making an explicit statement beforehand increases the chances that they will have to resign if and when the peg is abandoned. But making the promise is a way of buying a bit of credibility, and buying some time. Specifically, it is a device for signaling that their determination to hold the line on the currency is so strong that they are willing to risk sacrificing their jobs.

We selected a subsample of 24 cases out of our total set of currency crashes. We chose roughly equal numbers of cases with and without subsequent premier changes.¹¹ We searched local newspapers for the 30 days preceding the devaluation for statements by government officials that could be construed as commitments not to devalue. We included assurances even if the language did not read as explicit or ironclad, because these are so often interpreted as promises.¹²

The sample size was small. But we found that when a member of the government (chief executive, finance minister, or central bank governor) gave assurances that there would be no devaluation and yet a devaluation did subsequently occur, the probability that the chief executive would lose his or her job within 12 months was 2/3. When no such assurances were reported, the frequency of job loss was only 7/18, despite the devaluation. In other words, whatever the credibility benefits of the promise ex ante, it almost doubles the likelihood that the leader loses office ex post. If we use the six-month horizon, then the relative effect is even stronger: the leader is more than twice as likely to be out on the street if the government had made a previous commitment than if it had remained quiet (0.50 versus 0.19). If we consider only cases where the chief executive is the one to have given the assurances, then the job-loss rate becomes 100 percent. But there were only 2 such cases, out of 24. Usually the dangerous task of making assurances is delegated to a cabinet member. (Details are reported in Appendix 3 for the 12-month horizon and in Appendix 4 for the 6-month horizon.)

Despite this suggestive outcome, to the effect that the “broken-promise” factor does indeed matter, it seems unlikely that the “broken-promise” effect is the sole reason for devaluations to result in turnover at the top. After all, even among those cases where our newspaper search turned up no record of assurances in the month preceding the devaluation—either from the leader, finance minister, or central bank governor—22 percent of the leaders lost office within 6 months of the devaluation anyway and 39 percent lost within 12 months. These percentages are well above the 11.6 percent and 20.5 percent rates, respectively, of job loss in normal times. Evidently, the economic effects of devaluation also play an important role.

The Decline in Exchange Rate Pass-Through in Developing Countries

Conventional wisdom has long been that pass-through is slower or less complete in large industrialized countries than in small developing countries. A number of authors have pointed out a further decline during the 1990s in the pass-through coefficient among industrialized countries. But most of the many econometric studies of pass-through, even those that examine a recent decline in the pass-through coefficient, have focused on prices of imports into industrialized countries, rather than into developing countries. Taylor (2000) proposed that a decline in pass-through of exchange rate changes into the CPI in the 1990s was due to a lower inflationary environment and looked at U.S. data. Gagnon and Ihrig (2004) extended this claim to a sample of 11 industrialized countries. Otani, Shiratsuka, and Shirota (2003) found a similar decline in pass-through for imports into Japan. Campa and Goldberg (2002) again found a decline in the coefficient in the 1990s but attributed it more to changing commodity composition than to a less inflationary environment.²⁰ Their data set also consisted solely of industrialized countries.

Only a few studies include lower-income countries. Choudhri and Hakura (2001) extend to a sample of 71 countries, including developing countries, the finding that a low-inflation environment reduced pass-through to the CPI in the 1990s. Borensztein and De Gregorio (1999) and Goldfajn and Werlang (2000) study the low pass-through of recent large devaluations in developing countries.²¹ But these are all studies of influences on aggregate price measures, the CPI in particular, not on import prices. Few studies concentrate on imports of specific goods into developing countries. The difference is important because effects on price indices versus prices of specific imports are really two distinct conceptions of the word “pass-through.” It is even more important because, as in the rich-country context, some authors have claimed that what appears to be slow or incomplete pass-through in developing countries can really be attributed to changes in composition with regard to product varieties.²²

Table 1, taken from Frankel, Parsley, and Wei (2005), reports estimates for pass-through to prices of narrowly defined retail imports into 76 countries. Notice, first, confirmation of the conventional wisdom that pass-through has historically been higher in developing countries than in rich countries. As of the beginning of our sample period, 1990, the coefficient was 0.3 for rich countries and 0.8 for developing countries, with the difference highly significant statistically. (Figure 2, which illustrates the numbers on average during our sample period.) That these numbers fall below 1.0 cannot be attributed to compositional effects, as the eight goods are defined very narrowly: a roll of color film, a carton of Marlboro cigarettes, an issue of Time magazine, a bottle of Cointreau, and so forth.

Table 1.

Determination of Pass-Through to Imported Goods Prices: Developing Countries Relative to Rich Countries (76 countries, 1990–2001)

Source: Frankel, Parsley, and Wei (2005). Notes: Dependent variable: change in retail import prices of eight narrowly defined commodities. The eight import commodities (given with their country of origin) are: Marlboro cigarettes (U.S.), Coca-Cola (U.S.), cognac (France), Gilbey’s gin (U.S.), Time magazine (U.S.), Kodak color film (U.S.), Cointreau liqueur (France), and Martini & Rossi vermouth (Italy). Levels of significance are 5 percent (^**) and 1 percent (^***). For developing country coefficients, values in the “A Dev.” column can be added to those in the “Rich” column.

Table 1.

Determination of Pass-Through to Imported Goods Prices: Developing Countries Relative to Rich Countries (76 countries, 1990–2001)

Estimated Coefficient on		Rich	Δ Dev.
	Change in exchange rate	0.310***	0.496***
		(0.075)	(0.101)
	Change in exporter’s price	0.108***	-0.023
		(0.025)	(0.042)
	(Change in exchange rate) X trend	-0.025***	-0.026**
		(0.009)	(0.013)
Error correction term (ECM)		-0.091***	-0.105***
		(0.016)	(0.025)
ECM X trend		0.000	0.011***
		(0.002)	(0.003)
Number of observations		5,677
Adjusted R-squared		0.324

Source: Frankel, Parsley, and Wei (2005). Notes: Dependent variable: change in retail import prices of eight narrowly defined commodities. The eight import commodities (given with their country of origin) are: Marlboro cigarettes (U.S.), Coca-Cola (U.S.), cognac (France), Gilbey’s gin (U.S.), Time magazine (U.S.), Kodak color film (U.S.), Cointreau liqueur (France), and Martini & Rossi vermouth (Italy). Levels of significance are 5 percent (^**) and 1 percent (^***). For developing country coefficients, values in the “A Dev.” column can be added to those in the “Rich” column.

View Table

Table 1.

Determination of Pass-Through to Imported Goods Prices: Developing Countries Relative to Rich Countries (76 countries, 1990–2001)

Estimated Coefficient on		Rich	Δ Dev.
	Change in exchange rate	0.310***	0.496***
		(0.075)	(0.101)
	Change in exporter’s price	0.108***	-0.023
		(0.025)	(0.042)
	(Change in exchange rate) X trend	-0.025***	-0.026**
		(0.009)	(0.013)
Error correction term (ECM)		-0.091***	-0.105***
		(0.016)	(0.025)
ECM X trend		0.000	0.011***
		(0.002)	(0.003)
Number of observations		5,677
Adjusted R-squared		0.324

Source: Frankel, Parsley, and Wei (2005). Notes: Dependent variable: change in retail import prices of eight narrowly defined commodities. The eight import commodities (given with their country of origin) are: Marlboro cigarettes (U.S.), Coca-Cola (U.S.), cognac (France), Gilbey’s gin (U.S.), Time magazine (U.S.), Kodak color film (U.S.), Cointreau liqueur (France), and Martini & Rossi vermouth (Italy). Levels of significance are 5 percent (^**) and 1 percent (^***). For developing country coefficients, values in the “A Dev.” column can be added to those in the “Rich” column.

View Full Size

Pass-Through and Income, Average 1990–2001

(Country grouping based on World Bank classification)

Citation: IMF Staff Papers 2005, 002; 10.5089/9781589064485.024.A001

Source: Frankel, Parsley, and Wei (2005)—prices for 8 narrowly defined commodities imported into 76 countries; effect of exchange rate change within one year.Note: Pass-through for less developed countries is greater than for rich countries, historically.

• Download Figure
• Download figure as PowerPoint slide

Theories of slow or incomplete pass-through can be divided according to what sort of arbitrage barrier they posit as blocking the enforcement of the law of one price: barriers to international trade such as tariffs and transportation costs, or local costs of distribution and retail. The results in Frankel, Parsley, and Wei (2005) support both theories. Bilateral distance is a statistically significant determinant of the Error Correction Mechanism (ECM) term; that is, higher transport costs lead to slower pass-through to import prices. At the same time, a higher wage—the largest component of local distribution and retail costs—also shows up as a significant negative determinant of the pass-through coefficient.²³ Both determinants apply to rich and poor countries alike. Size does not appear as a determinant in most of our results: small countries do not experience more pass-through than do large ones, a very surprising finding in light of “pricing to market” theories (that is, price discrimination by sellers).

For present purposes, the important points are that the pass-through coefficient fell significantly in the 1990s and that the speed of decline was twice as fast among developing countries as it was among rich ones (0.051 per year compared with 0.025).²⁴ The speed of pass-through, which is estimated in the form of an ECM term, also shows a significant downward trend for developing (not for rich) countries.

One might wonder if this estimated decline in the pass-through coefficient during the 1990s is an indirect reflection of an asymmetry whereby pass-through of depreciation is greater than pass-through of appreciation, or a threshold effect whereby large devaluations result in proportionately less pass-through. We have found in extensions that the answer is “no”; the trend remains even after controlling for the big devaluations.²⁵

One would expect pass-through to prices of domestically produced goods or the general CPI to be (even) lower than to prices of imports. Our paper also reports results for other local price measures, and this is indeed the pattern they show (Figure 3). Tariffs and distance both contribute significantly to low pass-through to the CPI. But pass-through to prices of local substitutes and to the CPI both show the same downward trends over the sample period as does pass-through to import prices. The difference in coefficient trends between poor and rich countries is even greater for pass-through to the CPI than it is for import prices. This is important, in the present context, because most of the potential contractionary effects of devaluation require that pass-through extend beyond import prices alone, to include pass-through to locally produced goods or the CPI.

View Full Size

Exchange Rate Pass-Through to Domestic Prices

Citation: IMF Staff Papers 2005, 002; 10.5089/9781589064485.024.A001

Source: Frankel, Parsley, and Wei (2005)—effect within one year, in 76 countries.Note: Pass-through is greatest for prices of imports at dock, but less for retail and CPI.

• Download Figure
• Download figure as PowerPoint slide

What can we say about the reasons for the decline in pass-through? As noted, one hypothesis proposed by others is declining long-run inflation rates. This factor turns out to be particularly relevant in the case of explaining the downward trend in pass-through to developing country CPIs.

Another possible explanation for the trend is rising labor costs in retail and distribution. W e find that wages are a significant determinant of the pass-through coefficient. The wage hypothesis turns out to be particularly relevant in the explaining the downward trends in pass-through either to the prices of local substitutes or to the CPI. Controlling for wages reverses an estimated tendency for pass through to the CPI to decline as a country’s per capita income grows. A possible interpretation is that the role of distribution and retail costs in pricing to market becomes increasingly important as countries achieve higher incomes, owing to the Balassa-Samuelson-Baumol effect.

In any case, most of the decline in pass-through remains unexplained, despite the many contributing factors we estimated²⁶. The strongest conclusion is simply that incomplete pass-through is another respect in which developing countries have become a bit more like rich countries, for whatever reason.

The Balance Sheet Effect

If the contractionary effects that rely on pass-through to higher goods prices do not Explain the recessions that followed many of the 1990s devaluations, then what does? On the list of contractionary channels, the balance sheet effect is the one that has dominated in terms of attention from researchers, appropriately so. Domestic banks and firms had large debts denominated in foreign currencies, particularly in dollars. They might have been able to service their debt at the previous exchange rate, but they had trouble servicing it after the price of foreign exchange had gone up sharply. The results were layoffs and bankruptcies.²⁷

There is plenty of evidence of the output cost associated with the balance sheet effect. Looking at the experience of the 1990s, Cavallo and others (2002) show that countries entering a crisis with high levels of foreign debt tend to experience large real exchange rate overshooting (devaluation in addition to the long-run equilibrium level) and large output contractions. Similarly, Guidotti, Sturzenegger, and Villar (2004) find evidence that liability dollarization worsens output recovery after a sudden stop in capital inflows. Céspedes (2004) finds that the interaction of real devaluation and external debt has a significant negative effect on output.

It is easier to point out the problem of “mismatch”—between the currency of denomination of a country’s debts and the currency its firms earn—than it is to identify a cause, let alone a remedy. It is not enough to instruct firms to avoid dollar debts or to hedge them, because international investors are not very interested in lending to these countries in their own currencies. The result of following a rule to avoid borrowing in foreign currency would thus be to borrow less in total (which, admittedly, might not be such a bad outcome). Eichengreen and Hausmann (1999) have made the inability to borrow in local currencies famous under the name “original sin.” The phrase is meant to imply that the problem is not the fault of the countries themselves, or at least not the fault of recent governments. But we need not accept that it is completely predetermined.²⁸

Shifts on the Balance Sheet During the “Procrastination Phase”

The short-run question over which countries have some control arises during the interval I have called the period of procrastination. When foreign investors lose their enthusiasm for financing a country’s current account deficit, the national policymakers must decide whether to adjust or to wait. Typically, they wait. Countries that had managed to keep dollar-denominated debt relatively low tended to switch the composition of their debt in that direction during the year or so preceding the ultimate currency crash, to entice skeptical foreign investors to stay in.

A prime example is Mexico in 1994. International enthusiasm for investing in Mexico began to decline after the beginning of the year. The authorities clung to the exchange rate target and delayed adjustment, hoping circumstances would turn around. Most obviously, during much of the year they ran down reserves, as shown in Figure 4. But an important second mechanism of delay was to placate nervous investors by offering them tesobonos (short-term, dollar-linked bonds) in place of the peso bonds (Cetes) they had previously held. Figure 5 shows the dramatic increase in dollar-linked debt during the year leading up to the peso crisis of December 1994. It seems likely that the magnitude of the Mexican recession in 1995 stemmed not just from the adverse balance sheet effects that have been so frequently noted, but particularly from the adverse shift in balance sheets that took place in 1994. A third mechanism of delay was a shift toward shorter maturities, illustrated in Figure 6.²⁹ And the fourth has already been noted, an explicit commitment to defend the peg.

View Full Size

Evolution of Mexico’s Reserves, from Sudden Stop to 1994 Currency Crash, January 1992–April 1995

In billions of U.S. dollars

Citation: IMF Staff Papers 2005, 002; 10.5089/9781589064485.024.A001

Source: IMF, International Financial Statistics.

• Download Figure
• Download figure as PowerPoint slide

View Full Size

Evolution of Mexican Debt According to Currency Denomination, 1992–95

In percent

Citation: IMF Staff Papers 2005, 002; 10.5089/9781589064485.024.A001

Source Mexican Ministry of Finance and Public Credit.

• Download Figure
• Download figure as PowerPoint slide

View Full Size

Evolution of Mexican Debt According to Maturity, 1992–95

In percent

Citation: IMF Staff Papers 2005, 002; 10.5089/9781589064485.024.A001

Source Mexican Ministry of Finance and Public Credit.

• Download Figure
• Download figure as PowerPoint slide

These mechanisms are part of a strategy that is sometimes called “gambling for resurrection.” What they have in common, beyond achieving the desired delay, is helping to make the crisis worse when it does come, if it comes.³⁰ It is harder to restore confidence after a devaluation if reserves are near zero and the ministers have lost personal credibility. Further, if the composition of the debt has shifted toward the short term, in maturity, and toward the dollar, in denomination, then restoring external balance is likely to wreak havoc with private balance sheets, regardless of the combination of increases in interest rate versus increases in exchange rate.

The lesson? Adjust sooner rather than later (which is, admittedly, easier said than done).

Openness to Trade Reduces Vulnerability to Currency Crises

One final question concerns an aspect of the structure of the economy that can be influenced by policy, but only in the long run: the degree of integration with respect to international trade. Broadly speaking, there are two opposing views on the relationship between a country’s openness and whether it is prone to sudden stops or currency crashes. The first view is that openness makes a country more vulnerable to sudden stops. A country highly integrated into world markets is more exposed to shocks coming from abroad. The second view is that countries that are open to international trade are less vulnerable to sudden stops. If the ratio of trade to GDP is structurally high, it is easier to adjust to a cutoff in international financing of a given magnitude. I will describe a new test of the relationship between trade openness and vulnerability to sudden stops to help choose between the two hypotheses. Such tests have been done before, but usually without taking into account the possible endogeneity of trade. The incremental contribution here is to use the gravity instrument for trade openness—which aggregates geographically determined bilateral trade across a country’s partners—to correct for the possible endogeneity of trade.

The view that trade openness makes countries more vulnerable to crises comes in a number of forms. One variant is that a weakening in a country’s export markets is sometimes the trigger for a sudden stop in capital flows, so that a high-trade country is more vulnerable. Another variant notes that sudden stops in finance often extend to a loss in trade credit—especially for imports, but sometimes even for exports—and that the resulting shrinkage in trade is more painful if trade was a larger share of the economy to begin with. A third variant says that financial openness raises vulnerability to sudden stops, and openness to trade in practice goes hand in hand with openness to financial flows.³¹ In the limiting case, a country that is in autarky with respect to trade must have a net capital account of zero owing to the balance of payments adding-up constraint. Regardless of the specific reasoning, the notion that globalization leads to crises is a generalization that appeals to many.

The view that openness to trade makes countries less vulnerable also comes with a number of different specific mechanisms that have been proposed. Eaton and Gersovitz (1981) and Rose (2002) argue that the threatened penalty of lost trade is precisely the answer to the riddle, “Why do countries so seldom default on their international debts?” Strong trade links are statistically correlated with low default probabilities. International investors will be less likely to pull out of a country with a high trade-to-GDP ratio, because they know the country is less likely to default. A higher ratio of trade is a form of “giving hostages” that makes a lending cutoff less likely.

Another variant of the argument that openness reduces vulnerability takes as the relevant penalty in a crisis the domestic cost of adjustment, that is, the difficulty of eliminating a newly unfinanceable trade deficit. The argument goes back at least to Sachs (1985, p. 548). He suggested that Asian countries were less vulnerable to dislocations than Latin American countries in the international debt crisis of the 1980s—despite similar debt-to-GDP ratios—because they had higher export-to-GDP ratios. The relatively worse preformance observed in Latin America was due to the lower availability of export revenue to service debt. More recently, Guidotti, Sturzenegger, and Villar (2004) make a similar point by providing evidence that economies that trade more recover fairly quickly from the output contraction that usually comes with the sudden stop, while countries that are more closed suffer sharper output contraction and a slower recovery.

Consider first a country that faces a given cutoff in financing and must adjust without nominal or real exchange rate flexibility. The adjustment must then come through a reduction in spending. To achieve a $1 billion improvement in the trade balance, the contraction has to be $ (1/m) billion, where m is defined as the marginal propensity to import (in a Keynesian model) or the share of spending that falls on tradable goods (in a tradable/nontradable model). The lower m is, the more painful the adjustment. Whether output itself falls depends, of course, primarily on whether wages and prices are flexible. But even in a full-employment world, sharp reductions in consumption are not enjoyable.

Consider, second, a country that does have the option of nominal and real exchange rate flexibility. In traditional textbook models, if the adjustment is achieved in part through nominal and real depreciation, rather than exclusively through expenditure reduction, the country can accommodate the tougher new financing constraint without necessarily suffering a recession. This is true even if a relatively large devaluation is required to generate the necessary improvement in the trade balance. But since the emerging market crises of 1994–98, as we have already noted, economists have increasingly emphasized the contractionary balance sheet effect: if a country’s debts are denominated in foreign currency, the balance sheets of the indebted banks and corporations are hit in proportion to the devaluation. If the economy is starting from a high ratio of trade to GDP, the necessary devaluation need not be large, and therefore the adverse balance sheet effect need not be large. But if the economy is not very open to trade to begin with, the necessary devaluation, and the resulting balance sheet impact and recession, will be large. Again we arrive at the result that whether the necessary adjustment will be large and painful depends inversely on openness.

The balance sheet version of the openness story is modeled formally by Calvo, Izquierdo, and Talvi (2003) and Cavallo (2004). Both have in mind the example of Argentina, which has traditionally had a low ratio of trade to GDP and has suffered some of the worst sudden stops.³² But the hypothesis that openness to trade reduces a country’s vulnerability to sudden stops transcends any one formal model, causal link, or country example. The same is true of the opposing hypothesis, that openness raises a country’s vulnerability. It would be useful to be able to choose empirically between the two competing hypotheses.

I will report new results—from Cavallo and Frankel (2004)—for two questions. (1) What is the effect of openness on vulnerability to sudden stops implemented by a probit model measuring the probability of a sudden reduction in the magnitude of net capital inflows, following closely the definition of Calvo, Izquierdo, and Mejía (2004)?³³ (2) What is the effect of openness on vulnerability to currency crises, implemented by a probit model representing the probability of a sudden increase in exchange market pressure, which is in turn defined as the percentage depreciation plus percentage loss in foreign exchange reserves? In addition to analyzing the probit model of this exchange market pressure definition of a currency crisis, as in Frankel and Wei (2005), we also looked at the output loss subsequent to a crisis.

There is no reason, a priori, why something (openness) that makes the consequences of sudden stops better (less contractionary devaluations) should also necessarily make them less frequent, or that something that makes the consequences worse should also make them more frequent. Indeed, some theories are based on the notion that the worse the consequences, the less often it will happen. But in our results the effects turn out to go the same way, regardless of which concept of performance is used.

Calvo, Izquierdo, and Mejía (2004) and Edwards (2004a and b) are among the empirical papers that find that openness to trade is associated with fewer sudden stops. On the other hand, Milesi-Ferretti and Razin (1998 and 2000) find that openness helps trigger crises and/or sharp reversals of the current account. Most of these papers use the trade-to-GDP ratio as the measure for openness to trade.

A critic might argue that the trade-to-GDP ratio is endogenous. One way in which trade openness could be endogenous is via income: countries tend to liberalize trade barriers as they grow richer—in part because their mode of public finance gradually shifts from tariff revenue to income taxes or Value Added Tax (VAT). A second way is that trade liberalization could be part of a more general reform strategy driven by a proglobalization philosophy or “Washington Consensus” forces. Other aspects of such a reform program, such as privatization, financial liberalization, or macroeconomic stabilization might affect the probability of crises, and yet an Ordinary Least Squares regression analysis (OLS) might inappropriately attribute the effect to trade. A third way that trade openness could be endogenous is that experience with crises—the dependent variable—may itself cause liberalization, via an IMF program. Or it might have the opposite effect, if a country’s response to a crash is disenchantment with globalization and the Washington Consensus. A fourth way in which trade openness could be endogenous is through the feedback between trade and financial openness.

How can the endogeneity of trade be addressed? We use gravity estimates to construct an instrumental variable for trade openness, a methodology developed by Frankel and Romer (1999), in the context of the effect of trade on growth, and updated in the Frankel and Rose (2002) data set.

The results reported in Table 2 show that openness reduces vulnerability to sudden stops rather than increasing it.³⁴ Not only does this relationship hold up when we move from OLS to instrumental variables, but it appears stronger. The degree of trade openness is a powerful predictor of these capital account shocks: moving from Argentina’s current trade share (approximately .20 of GDP) to Australia’s average trade share (approximately .30 of GDP) reduces the probability of a sudden stop by 32 percent. The results for openness are the same when we seek to explain currency crashes. Trade protectionism does not shield countries from the volatility of world markets, as proponents might hope. On the contrary, less trade openness leads to greater vulnerability to sudden stops and currency crashes. In fact, out of the set of controls we tried, openness is the only variable that is virtually always statistically significant.³⁵

Table 2.

Effect of Openness (Trade/GDP) on Vulnerability to Sudden Stops and Currency Crashes

Source: Cavallo and Frankel (2004) Notes: Robust standard errors reported in parentheses. ^*** = statistically significant at 1 percent. ^** = statistically significant at 5 percent. ^* = statistically significant at 10 percent. Estimation performed with regional dummies and year fixed effects. IV is the gravity-based instrumental variable for trade openness from Frankel and Romer (1999) and Frankel and Rose (2002).

¹Calvo, Izquierdo, and Talvi (2003), definition.

²Frankel and Wei (2005), definition.

Table 2.

Effect of Openness (Trade/GDP) on Vulnerability to Sudden Stops and Currency Crashes

	To Predict Sudden Stops1		To Predict Currency Crashes2
	Ordinary probit	IV	Ordinary probit	IV
Trade opennesst	−0.53	−2.45	−0.57	−1.73
	(0.259)**	(0.813)**	(0.269)**	(0.918)*
Foreign debt/GDPt-1	−0.080	0.196	0.23	0.59
	(0.217)	(0.275)	(0.231)	(0.373)
Liability dollarizationt-1	0.316	0.591	0.027	0.18
	(0.195)	(0.256)**	(0.249)	(0.234)
Exchange rate rigidity			0.13	0.22
			(0.094)	(0.113)*
CA/GDPt-1	−4.068	−7.386	−0.272	0.66
	(1.297)**	(2.06)***	(1.392)	(1.455)
ln reserves in months of imports t-1			−0.26	−0.37
			(0.082)***	(0.099)***
Constant	−2.544	−1.73	−0.99	0.304
	(0.63)***	(0.723)**	(0.749)	(0.786)
Number of Observations	778	1,062	557	841

Source: Cavallo and Frankel (2004) Notes: Robust standard errors reported in parentheses. ^*** = statistically significant at 1 percent. ^** = statistically significant at 5 percent. ^* = statistically significant at 10 percent. Estimation performed with regional dummies and year fixed effects. IV is the gravity-based instrumental variable for trade openness from Frankel and Romer (1999) and Frankel and Rose (2002).

¹Calvo, Izquierdo, and Talvi (2003), definition.

²Frankel and Wei (2005), definition.

View Table

Table 2.

Effect of Openness (Trade/GDP) on Vulnerability to Sudden Stops and Currency Crashes

	To Predict Sudden Stops1		To Predict Currency Crashes2
	Ordinary probit	IV	Ordinary probit	IV
Trade opennesst	−0.53	−2.45	−0.57	−1.73
	(0.259)**	(0.813)**	(0.269)**	(0.918)*
Foreign debt/GDPt-1	−0.080	0.196	0.23	0.59
	(0.217)	(0.275)	(0.231)	(0.373)
Liability dollarizationt-1	0.316	0.591	0.027	0.18
	(0.195)	(0.256)**	(0.249)	(0.234)
Exchange rate rigidity			0.13	0.22
			(0.094)	(0.113)*
CA/GDPt-1	−4.068	−7.386	−0.272	0.66
	(1.297)**	(2.06)***	(1.392)	(1.455)
ln reserves in months of imports t-1			−0.26	−0.37
			(0.082)***	(0.099)***
Constant	−2.544	−1.73	−0.99	0.304
	(0.63)***	(0.723)**	(0.749)	(0.786)
Number of Observations	778	1,062	557	841

Source: Cavallo and Frankel (2004) Notes: Robust standard errors reported in parentheses. ^*** = statistically significant at 1 percent. ^** = statistically significant at 5 percent. ^* = statistically significant at 10 percent. Estimation performed with regional dummies and year fixed effects. IV is the gravity-based instrumental variable for trade openness from Frankel and Romer (1999) and Frankel and Rose (2002).

¹Calvo, Izquierdo, and Talvi (2003), definition.

²Frankel and Wei (2005), definition.