Exchange Rate Management and Crisis Susceptibility: A Reassessment†
Author: Mr. Atish R. Ghosh, Mr. Jonathan David Ostry, and Miss Mahvash S Qureshi

Contributor Notes

Authors’ E-Mail Addresses: aghosh@imf.org; jostry@imf.org; mqureshi@imf.org
Publication Date:
24 Jan 2014
eISBN:
9781484383971
Language:
English
Keywords:
WP; solid windowtext 1pt; intermediate exchange rate regime; currency crisis; banking crisis; regime classification; managed float category; crisis susceptibility; crisis observation
Download PDF (1.6 MB)

This paper revisits the bipolar prescription for exchange rate regime choice and asks two questions: are the poles of hard pegs and pure floats still safer than the middle? And where to draw the line between safe floats and risky intermediate regimes? Our findings, based on a sample of 50 EMEs over 1980-2011, show that macroeconomic and financial vulnerabilities are significantly greater under less flexible intermediate regimes—including hard pegs—as compared to floats. While not especially susceptible to banking or currency crises, hard pegs are significantly more prone to growth collapses, suggesting that the security of the hard end of the prescription is largely illusory. Intermediate regimes as a class are the most susceptible to crises, but “managed floats”—a subclass within such regimes—behave much more like pure floats, with significantly lower risks and fewer crises. “Managed floating,” however, is a nebulous concept; a characterization of more crisis prone regimes suggests no simple dividing line between safe floats and risky intermediate regimes.

Abstract

This paper revisits the bipolar prescription for exchange rate regime choice and asks two questions: are the poles of hard pegs and pure floats still safer than the middle? And where to draw the line between safe floats and risky intermediate regimes? Our findings, based on a sample of 50 EMEs over 1980-2011, show that macroeconomic and financial vulnerabilities are significantly greater under less flexible intermediate regimes—including hard pegs—as compared to floats. While not especially susceptible to banking or currency crises, hard pegs are significantly more prone to growth collapses, suggesting that the security of the hard end of the prescription is largely illusory. Intermediate regimes as a class are the most susceptible to crises, but “managed floats”—a subclass within such regimes—behave much more like pure floats, with significantly lower risks and fewer crises. “Managed floating,” however, is a nebulous concept; a characterization of more crisis prone regimes suggests no simple dividing line between safe floats and risky intermediate regimes.

I. Introduction

“Whatever exchange rate system a country has, it will wish at some times that it had another one.” - Stanley Fischer (1999)

The choice of exchange rate regime is a perennial question facing emerging market economies (EMEs). In a world of increasingly volatile capital flows, even if the ultimate decision depends on a variety of historical, political, and economic factors, any rational calculus on regime choice must take into account its crisis susceptibility. While a voluminous literature on regime vulnerabilities grew out of the EME crises of 1990s and early 2000s, the changing trends in regimes since then (most notably, toward managed floats), the large output drops experienced by EMEs under a variety of regimes during the global financial crisis (GFC), and more recently the impact of “tapering talk” on EME currencies, makes pertinent the question of which regimes are the most vulnerable to crisis, and why.

Conventional wisdom, articulated by Fischer (2001), is the bipolar prescription: countries should adopt floats or hard pegs (monetary union, dollarization, currency board) and avoid intermediate regimes, as they tend to be more susceptible to crisis. While the arguments in favor of free floats are well known, it is less clear why hard pegs—the least flexible regime—should be equally resilient to crisis. Certainly the experience of emerging Europe and some eurozone countries during the GFC suggests that hard pegs may be more prone to growth declines and painful current account reversals, in which case the hard end of the bipolar prescription may be largely illusory.

But the soft end of the prescription is also problematic, a key question there being where to draw the line between floats and more risky intermediate exchange rate regimes. Clearly, occasional interventions during periods of market turbulence or extreme events do not turn a float into an intermediate regime; but how much management of the exchange rate is too much? This is the policy question confronting many EME central banks, an increasing number of which have switched to “managed floats”—i.e., regimes where the central bank does not (at least explicitly) target a particular parity—as they decide in real time how (or whether) to respond to various shocks. Even central banks intending to float freely may find themselves straying toward increased management of the exchange rate as they react to unfolding events, in turn generating expectations that a de facto intermediate exchange rate regime is in place.

The existing literature provides limited, and generally contradictory, guidance on how much management of the exchange rate is too much. In his seminal work, Fischer (2001, 2008) put “managed floats” with free floats—that is, at the safe pole—rather than with the risky intermediate regimes. More generally, for countries with open capital accounts, he considers “a wide range of arrangements running from free floating to a variety of crawling bands with wide ranges” to be appropriate. But most other studies (e.g., Eichengreen, 1994; Obstfeld and Rogoff, 1995; Frankel, 1999; Masson, 2000; Rogoff et al., 2004), adopt a more extreme version of the bipolar view—lumping managed floats (or regimes with wide bands) with other intermediate exchange rate regimes. Rogoff et al. (2004), for example, find that managed floats are significantly more prone to financial crisis than free floats, arguing that EMEs would benefit from “learning to float.” And in the context of the broad-band regime (±15 percent around a central rate) adopted by the European Monetary System (EMS) after the crisis of 1992-93, Obstfeld and Rogoff (1995) argue that such systems pose difficulties, and that “there is little, if any, comfortable middle ground between floating rates and the adoption by countries of a common currency.”

In this paper, therefore, we examine two related questions: Does the bipolar prescription still hold in the sense that the extremes are safer than the middle? And, at the flexible end, where to draw the line between safe floats and risky intermediate regimes? For our analysis, we go beyond the usual three-way fixed, intermediate and float categorization, and adopt the IMF’s detailed de facto classification—which allows us to differentiate among the various intermediate exchange rate regimes—and supplement it with other popular classifications (such as the IMF’s de jure and Reinhart and Rogoff’s (2004) de facto classifications). For each regime, we examine the underlying vulnerabilities (macroeconomic imbalances, financial-stability risks) and the frequency of banking, currency, and sovereign debt crises. Ultimately, however, we are interested in crises because of their impact on welfare, the simplest yardstick of which is output growth. While growth indeed declines sharply during these crises, it is possible that certain regimes are associated with growth collapses that are independent of—or at least not manifested in—one of these crises. To address this possibility, we round out our crisis definitions by adding growth collapses—i.e., sharp decelerations of growth relative to the country’s historical norm.

Turning to the line between safe floats and risky intermediate regimes, we find that different regime classifications yield sharply different results for the “managed float” category. We therefore need to go beyond “canned” classifications and instead identify the more crisis prone regimes in terms of their primitive characteristics such as nominal exchange rate flexibility (across various horizons) and degree, direction, and circumstances of foreign currency (FX) intervention.1 For this purpose, we adopt an innovative decision-theoretic technique, known as binary recursive tree (BRT) analysis, which allows for arbitrary thresholds and interactive effects among the explanatory variables (e.g., exchange rate flexibility; degree of FX intervention; overvaluation of the real exchange rate, etc.) in determining crisis susceptibility.

Our results, based on a dataset of 50 major EMEs over 1980-2011, could be summarized as follows. First, when it comes to financial vulnerabilities (rapid credit expansion; excessive foreign borrowing; FX-denominated domestic currency lending), and macroeconomic vulnerabilities (currency overvaluation; delayed external adjustment), less flexible intermediate regimes (pegs, bands, and crawls) are significantly more vulnerable than pure floats—but so are hard pegs. Second, intermediate exchange rate regimes as a class are indeed the most susceptible to banking and currency crisis, but de facto managed floats—a subclass within intermediate regimes—behave much more like pure floats, with significantly lower risks and fewer crises. The vulnerabilities under hard pegs however tend to be manifested in growth collapses rather than in banking or currency crises—perhaps because the high cost of exiting the regime makes the authorities reluctant to abandon it, opting instead for long and painful adjustment. Third, at the soft end, we find that there is no simple uni-dimensional dividing line (e.g., according to nominal exchange rate flexibility) between safe floats and risky intermediate regimes. Rather, the key to avoiding crises is to ensure that the real exchange rate does not become overvalued—and what makes for a “safe” managed float is that the central bank intervene in the face of overvaluation pressures and refrain from intervening to defend an overvalued exchange rate.

Our paper contributes to the existing literature in three respects. First, by going beyond existing studies and looking at underlying macroeconomic and financial vulnerabilities, we establish that the hard end of the bipolar spectrum is the most vulnerable—but that these vulnerabilities are manifested mainly in the form of growth crises, implying that the hard end of the bipolar prescription is largely illusory.2 Second, by using a finer exchange rate regime classification than the usual three-way categorization, we are able to establish that not all intermediate exchange rate regimes are alike: managed floats (as defined by the IMF’s de facto classification) are significantly less susceptible to banking crisis than basket pegs, crawls and bands, or to growth collapses than single currency and basket pegs. Third, by using the BRT analysis to get around the ambiguity across existing regime classifications, we are able to identify the more crisis-prone intermediate regimes according to such characteristics as the degree of nominal exchange rate flexibility and circumstances of FX intervention, which is likely to be more useful for policy purposes than how a canned classification categorizes the regime.

The remainder of this paper is structured as follows. Section II describes the IMF’s de facto regime classification used in the analysis, and documents the evolution of exchange rate regimes in EMEs over the past three decades. Section III briefly reviews why certain regimes may be more crisis-prone, and then examines the empirical evidence on their vulnerabilities and susceptibility to banking, currency, sovereign debt, and growth crises. Section IV further explores the characteristics of more crisis-prone intermediate regimes through the use of binary recursive tree analysis. Section V concludes.

II. Trends in Exchange Rate Regimes in EMEs

As in any empirical study of exchange rate regimes, our first task is to choose a classification scheme. Early studies (e.g., Ghosh et al., 1995) used de jure classifications—the regime declared by the central bank—but since the bias of these classifications, whereby EME central banks typically claim to follow more flexible exchange rate arrangements than they actually do (characterized as “fear of floating;” Calvo and Reinhart, 2002), has become apparent, subsequent studies generally use de facto classifications (e.g., Ghosh et al., 2003; Reinhart and Rogoff, RR, 2004; Shambaugh, 2004; Levy-Yeyati and Sturzenegger, LYS, 2005).3 There is, however, little agreement among the various de facto classifications, and they often produce conflicting results in macroeconomic studies.4

Here we mainly use the IMF’s de facto classification, which combines statistical methods with qualitative judgment based on IMF country team analysis and consultations with the central bank.5 Compared to other de facto classifications, the IMF classification provides wider and more up-to-date coverage (including the period since the GFC), exhibits the greatest consensus across de facto classifications, and makes clearer distinctions between hard pegs and conventional pegs, and between managed floats and pure floats. Moreover, by combining (often confidential) information on the central bank’s intervention policy with actual exchange rate volatility, it avoids the occasional anomalies from which purely mechanical algorithms to classify regimes (as in other classifications) inevitably suffer.6

Three phases can be discerned in the evolution of EME exchange rate regimes over the past couple of decades (Figure 1 [a]). The 1997-98 Asian crisis, and its immediate aftermath, saw a “hollowing out of the middle”—countries abandoning single currency or other “soft” pegs (mostly in favor of free floats)—consistent with the bipolar prescription.7 This trend came to an end around 2004, however, with the proportion of intermediate exchange rate regimes rising in the runup to the GFC, mainly because of the increased adoption of managed floats by EMEs (Figure 1[b]). In the third phase, the GFC and beyond, the move toward intermediate exchange rate regimes, especially managed floats, has accelerated markedly.8

Figure 1.
Figure 1.

Distribution of Exchange Rate Regimes in EMEs: IMF’s De Facto Classification

1980-2011 (In percent)

Citation: IMF Working Papers 2014, 011; 10.5089/9781484383971.001.A001

The intervention patterns underlying the move to greater exchange rate management pre- and post-GFC are, however, quite different. In the runup to the crisis, most EMEs worried that capital inflows would make their exports uncompetitive and therefore sought to limit the appreciation of their currencies. During the crisis, when EMEs were facing a sudden stop or even sharp capital outflows, intervention was to support the currency. Thereafter, the ebbs and flows of capital to these countries have resulted in alternating phases of concern about currency appreciation and depreciation—but in any case, concern about exchange rate volatility, hence the desire to manage exchange rates.

But is this trend of EMEs moving toward managed floats likely to continue? If the bipolar view held as a positive prediction, then Markov transition matrices would imply that hard pegs and free floats would be absorbing states, and together form a closed set.9 Estimated transition probabilities for the full sample period (1980-2011) using the three-way classification show that this is not the case: while regimes tend to be highly persistent, none of the off-diagonal probabilities is zero, implying that transitions from every regime to another are possible (Table 1[a]). The floating regime is the least persistent, with many more transitions from floats to intermediate regimes (about 20 percent) than to hard pegs (2 percent). Consistent with earlier findings by Masson (2000) and Bubula and Ötker-Robe (2002), formal tests reject the bipolar view as a positive prediction (Table 1, last row). In fact, according to the steady-state distribution—assuming historical transition probabilities remain unchanged in the future and there are no major shocks to the system—intermediate regimes will be the most prevalent, with about 70 percent of EMEs opting for them in the long-run, and a further 20 percent opting for hard pegs.10 Similar results are obtained using a more recent sample (2000-11), except that it is not possible to reject the hypothesis that hard pegs are an absorbing state (though the bipolar prediction itself is strongly rejected).11

Table 1.

Transition Probabilities Matrix for EMEs: IMF’s De Facto Aggregate Classification

(a) 1980-2011

article image
Note: Fixed=hard pegs (no separate legal tender/currency board); Intermediate=pegs to single currency, basket pegs, horizontal band, crawling peg/band, and managed floats; Float=independent floats.

(b) 2000-2011

article image

Turning to transitions among intermediate regimes, Table 2 reports Markov matrices using the finer regime classification. In the full sample, basket pegs are the most persistent regime, followed by crawling pegs/bands, managed floats, and single currency pegs. In the more recent sample, however, while basket pegs remain the most persistent regime, crawls are less persistent, while managed floats and single currency pegs become more persistent. Exits from both hard pegs and floats are much more likely to be toward managed floats in both samples than to any other intermediate regime. The full-sample steady state distribution implies that managed floats would be the most dominant regime in the long-run, with a share of about 31 percent, followed by hard pegs (20 percent), crawling pegs/bands (17 percent), single currency pegs (14 percent) and floats (11 percent). Restricting estimation to the more recent sample suggests an equal split between hard pegs and managed floats of about 30 percent each, followed by single currency pegs (15 percent) and pure floats (10 percent).12

Table 2.

Transition Probabilities Matrix for EMEs: IMF’s De Facto Fine Classification

(a) 1980-2011

article image

(b) 2000-2011

article image

Overall, these findings strongly reject the bipolar hypothesis as a positive prediction. On the contrary, EMEs seem to favor intermediate exchange rate regimes—and managed floats in particular. Hard pegs may also be more prevalent as several emerging European countries are likely to join the eurozone. Determining the crisis risks of these regimes is therefore a pressing policy question.

III. Exchange Rate Regimes and Crisis Vulnerability

Underlying most crises is some form of vulnerability (unsustainable imbalances, excessive balance sheet exposures), and there are several reasons why these vulnerabilities would be worse under less flexible exchange rates regimes than under floats. First, the loss (or limit) of the exchange rate as an adjustment tool makes it more difficult to correct external imbalances, often to the point that the real exchange rate becomes overvalued and large imbalances build up, whose unwinding precipitates a currency crisis (often anticipated by a self-fulfilling speculative attack).13 Second, relatedly, regaining competitiveness without nominal exchange rate flexibility puts deflationary pressures on the economy, which in turn may undermine output growth. Third, the exchange rate guarantee implicit in the peg can encourage excessive foreign borrowing by banks (and other domestic entities), especially when there is a favorable interest rate differential for FX borrowing (Rosenberg and Tirpak, 2008; Magud et al., 2011). In turn, open FX limits on banks force them to lend in foreign currency, which is of particular concern when the ultimate borrowers (e.g., households) lack a natural FX hedge. Fourth, to the extent that intervention is not sterilized (e.g. due to the fiscal cost), there may be excessive credit expansion, exacerbated by the implicit exchange rate guarantee that attracts nonresident deposits and expands bank balance sheets (Montiel and Reinhart, 2001).14 Finally, by temporarily suppressing the effects of lax fiscal policy on inflation, less flexible exchange rate regimes may impose less fiscal discipline than flexible regimes (Tornell and Velasco, 2000).

The different types of vulnerabilities may also interact and amplify each other: sharp declines in growth can worsen debt sustainability and impair the quality of bank assets; greater foreign borrowing can lead to large exchange rate swings in the event of a sudden stop; but sharp currency movements can strain unhedged domestic balance sheets and result in growth slowdowns.15 But even if less flexible exchange rates are more likely to be vulnerable, the form of crisis in which the vulnerability is manifested may depend on the type of exchange rate regime. In particular, the high cost of exiting a hard peg—and therefore the policy discipline and market credibility engendered by it—makes currency crises less likely. The same features may also result in smaller fiscal deficits, and therefore, lower risk of debt sustainability problems under hard pegs (though by reducing the scope for inflationary finance, they may make discrete default more likely) than under other less flexible regimes.16

By contrast, the very determination of the authorities to maintain the parity means that growth crises are more likely (while the larger imbalances and exposures means that the output cost of any eventual currency crisis will be all the greater—as the collapse of Argentina’s currency board amply demonstrated). This suggests that in assessing the resilience of exchange rate regimes, it is important to go beyond the traditional currency and banking crises and also consider other types of crisis such as debt crises and growth collapses. Moreover, since crises are rare events (requiring both an underlying vulnerability and crisis trigger; see Ghosh et al., 2008), and may—serendipitously—not be realized in the sample, it is important to consider both underlying vulnerabilities and crisis realizations.17

A. Financial and Macroeconomic Vulnerabilities

We begin by examining the relationship between the exchange rate regime and various financial and macroeconomic vulnerabilities by estimating the following model:

Vji=xjt′β1+zjt′γ1+ηjt(1)

where Vjt is the financial vulnerability (rapid credit expansion; excessive foreign borrowing; FX-denominated lending) or macroeconomic vulnerability (fiscal and current account deficits; real exchange rate overvaluation) in country j in time t; x is a vector of binary variables indicating the exchange rate regime in place; z includes relevant control variables as identified in earlier literature; and η is the random error term. We estimate (1) using Ordinary Least Squares (OLS), and cluster the standard errors at the country level to address the possibility of correlation in the error term. To address potential endogeneity concerns of the exchange rate regime and control variables in (1), we follow existing literature (e.g., Rogoff et al., 2004) and substitute current values of these variables by lagged values. Since exchange rate regimes are slow moving variables, we do not include country-fixed effects, but control for region-specific effects and a range of country characteristics.18

Financial vulnerabilities

Empirical studies generally find that less flexible exchange rate arrangements are more likely to be associated with higher credit to the private sector (Magud et al., 2011) or credit booms (Mendoza and Terrones, 2008; Dell’Ariccia et al., 2012). The same is true in our data set, where change in domestic credit (defined as the 3-year cumulative change in the ratio of private sector credit-to-GDP) is almost twice as large under hard pegs as under intermediate regimes, and almost four times as large as under floats (Table 3, col. [1]). The aggregate statistic for intermediate regimes, however, masks important differences across these regimes: for instance, change in credit is more than twice as large under basket pegs than under single currency pegs—and almost eight times as large as under managed floats.

Table 3.

Vulnerabilities and Crisis in EMEs: IMF’s De Facto Classification, 1980-2011

(In percent)

article image

In percent of exchange rate regime observations.

In percentage points.

More formal regression analysis confirms these results: the change in credit-to-GDP ratio or credit expansion (i.e., restricting the sample to positive changes in credit-to-GDP ratio) is statistically significantly greater under hard pegs, single currency pegs, basket pegs, or horizontal bands than under pure floats (Table 4, cols. [2], [5]). While the control variables included in the estimation—based on earlier literature (e.g., Mendoza and Terrones, 2008; Magud et al., 2011)—such as real GDP growth, net capital inflows, and foreign borrowing by the banking system are all significant contributors to domestic credit expansion, the association between less flexible exchange rate regimes and private sector credit mostly survives their inclusion in the regression (cols. [3], [6]). Notably, change in credit/credit expansion under other less flexible intermediate regimes (single currency pegs, basket pegs, or horizontal bands) is also statistically significantly higher than under managed floats (as indicated by the test for coefficient equality reported in the last row, Table 4).

Table 4.

Domestic Credit: IMF’s De Facto Classification, 1980-2011

article image
Notes: Dependent variable is cumulative change in private sector credit to GDP ratio over 3 years (i.e., difference between t and t-3) in cols. [1]-[3]. Sample restricted to only positive changes in columns (4)-(6). The reference category is free float. All variables (except for initial credit/GDP ratio) are lagged one period. See Appendix B for variable definitions and data sources. Constant included in all specifications. Clustered standard errors at country level reported in parentheses. ***, ** and * indicate signficance at 1,5 and 10 percent levels, respectively.

As discussed above, the exchange rate guarantee implicit in a peg (or less flexible regimes more generally) might also encourage excessive foreign borrowing by the banking system and, given open FX limits, may imply a corresponding increase in FX-denominated lending to the private sector. The raw statistics reported in Table 3 (cols. [2]-[3]) suggest that this is indeed the case: both foreign borrowing (measured as foreign liabilities of the banking system, in percent of GDP) and FX-denominated domestic lending (share of domestic FX-denominated loans in total loans of the banking system) are twice as large under hard pegs as under floats, with intermediate regimes somewhat closer to the latter.

Regression analysis shows that foreign borrowing by the banking system is significantly greater under less flexible exchange rate regimes than under pure floats; and also under hard and single currency pegs as compared to managed floats (Table 5, col. [2]). These results generally continue to hold for hard pegs and single currency pegs when controlling for other explanatory variables (Table 5, cols. [3]-[6]). Hard pegs and basket pegs are also associated with a significantly greater proportion of FX-denominated lending in total bank lending as compared to free floats (Table 6, col. [2]), though the results weaken when we control for net capital flows and bank foreign liabilities, which suggests that less flexible regimes induce greater FX-denominated lending by encouraging funds from abroad (cols. [3]-[6]).19

Table 5.

Bank Foreign Borrowing: IMF’s De Facto Classification, 1980-2011

article image
Notes: Dependent variable is bank foreign liabilities to GDP (in percent). The reference category is free float. All variables are lagged one period. See Appendix B for variable definitions and data sources. Constant included in all specifications. Clustered standard errors at country level reported in parentheses. ***, ** and * indicate significance at 1,5 and 10 percent levels, respectively.
Table 6.

FX Lending: IMF’s De Facto Classification, 1995-2011

article image
Notes: Dependent variable is bank lending in FX to total bank lending (in percent). Dollarized countries are excluded from the estimation. The reference category is free float. All variables are lagged one period. The sample size drop as data on FX lending is available from 1995 onward. See Appendix B for variable definitions and data sources. Constant included in all specifications. Clustered standard errors at country level reported in parentheses. ***, ** and * indicate significance at 1,5 and 10 percent levels, respectively.

The regressions reported in Tables 5 and 6 also point to policy measures that can help reduce these risks. For instance, consistent with the findings of Ostry et al. (2012), controls on capital inflows are associated with significantly lower banking system external liabilities (Table 5, col. [4]) and, more surprisingly, with a lower proportion of FX-denominated domestic bank lending (Table 6, col. [4]).20 Likewise, restrictions on FX-denominated lending naturally reduce the proportion of such loans in total bank lending (Rosenberg and Tirpak, 2008; Ostry et al. 2012), while open FX-limits have a stronger impact on foreign borrowing by the banking system (Tables 5 and 6, cols. [5]-[6]).

Macroeconomic vulnerabilities

Beyond financial vulnerabilities, less flexible exchange rate regimes may be associated with greater macroeconomic vulnerabilities: fiscal deficits, current account deficits, and real exchange rate overvaluation. What is the formal empirical evidence? Fiscal deficits are lower under hard pegs than under most other less flexible exchange rate regimes—with the exception of basket pegs (Table 3, col. [4])—but the differences are not statistically significant from free floats (again, except for basket pegs, which have significantly higher fiscal balances than free floats; Table 7, cols. [2]-[3]). Both hard pegs and intermediate regimes are associated with significantly greater overvaluation of the real exchange rate (measured simply as the deviation of real effective exchange rate from trend) than pure floats—and this holds regardless of controlling for capital inflows (which itself is significantly associated with overvaluation). The fine classification, however, shows that it is hard pegs and single currency, basket, and crawling pegs that are susceptible to (statistically significant) overvaluation: managed floats do not exhibit greater overvaluation than pure floats (Table 7, cols. [4]-[6]).

Table 7.

Fiscal Balance and REER Deviation: IMF’s De Facto Classification, 1980-2011

article image
Notes: Dependent variable is general government fiscal balance to GDP (in percent) in cols. [1]-[3] and REER deviation from trend (in percent) in cols. [4]-[6]. The reference category is free float. All variables (except for terms of trade change) are lagged one period. Constant included in all specifications. Clustered standard errors at country level reported in parentheses. ***, ** and * indicate significance at 1, 5 and 10 percent levels, respectively.

Less flexible exchange rate regimes do appear to impede external adjustment—on average, current account imbalances tend to be larger under hard pegs and intermediate regimes than under floats (Figure 2). Prior to reversals—defined as large reductions in the current account imbalances—surpluses and deficits also tend to be larger under these regimes relative to pure floats (Table 8, cols. [1]-[3]). While nothing forces adjustment on surplus countries, deficit countries can lose financing abruptly especially when large imbalances have built up. Accordingly, the (unconditional) reversal probability is significantly greater for hard pegs and (almost) all intermediate exchange rate regimes (col. [4]).

Figure 2.
Figure 2.

Current Account Balance in EMEs: IMF’s De Facto Classification

1980-2011 (In percent)

Citation: IMF Working Papers 2014, 011; 10.5089/9781484383971.001.A001

Source: Anderson (2008), IMF’s AREAER and WEO databases.Note: The figure depicts the average surplus and deficit under different exchange rate regimes in our sample of EMEs. Thus, e.g., panel (a) shows that fixed, intermediate and floating regimes have, on average, current account deficits of -8, -6 and -4 percent of GDP, respectively; and current account surpluses of about 2.5, 4, and 2.5 percent of GDP, respectively
Table 8.

Current Account Reversals: IMF’s De Facto Classification, 1980-2011

(In percent of GDP)

article image
Note: Reversals defined as in Freund (2005). Prior balance indicates the maximum surplus or deficit prior to the reversal (in percent of GDP). Reversal probability indicates the frequency of reversal as a proportion of exchange rate regime observations. ***, ** and * indicate statistical significance of difference in proportions from the (independent) float category at the 1,5, and 10 percent levels, respectively.

B. Crisis Propensity

The findings above suggest that less flexible exchange rate regimes may be much more vulnerable to crisis. But do these vulnerabilities translate into actual crises? And are all less flexible regimes equally prone to different types of crisis? In this section, we empirically explore these questions with regard to banking, currency, and sovereign debt crises, as well as general growth collapses, by estimating models of the following form:

Pr⁡(Crisijt=1)F(xjt′β2+zjt′γ2)(2)

where Crisisjt is an indicator variable of whether a crisis (banking, currency, debt, or growth) occurs in country j in period t; x indicates the exchange rate regime in place (before the onset of the crisis), and z includes various relevant control variables (lagged). We estimate (2) using the probit model, and as before, include region-specific effects, and cluster the standard errors at the country level.

To define the various types of crisis, we follow the existing literature. Systemic banking crises are those where there are significant signs of financial distress in the banking system, requiring significant policy intervention methods in response to significant losses (Laeven and Valencia, 2012). Currency crises are depreciations of the nominal exchange rate against the US dollar of at least 30 percent that are also at least 10 percentage points greater than the depreciation in the previous year (Frankel and Rose, 1996). External debt crises are identified as events of sovereign debt default and/or restructuring.21 Growth collapses are defined as those that are in the bottom fifth percentile of growth declines (current year relative to the average of the three previous years), and correspond to a fall in the growth rate of real GDP of about 7.5 percentage points in our sample.

An initial snapshot shows that currency and banking crises are the most common form of financial crisis in EMEs, while sovereign debt crises are the least prevalent (Table 9). Large growth declines are also quite common, and not all of them are accompanied by another type of crisis. For example, only about one-third of growth collapses occur with or within three years of a banking or currency crisis (and fewer than 10 percent occur in the context of a debt crisis). Currency and debt crises seem to be strongly related to banking crisis—around one-half occur within three years of a banking crisis, whereas some 15-30 percent of banking crises happen within three years of a debt or currency crisis.

Table 9.

Crisis Occurrence in EMEs: 1980-2011

(In percent of crisis observations)

article image
Note: Table depicts the percentage of crisis observations preceded (in the last two years) or accompanied by other types of crises, along with the total number of crisis observations for each type of crisis.

Banking and currency crises

A number of studies have documented the higher propensity of banking crisis (Domaç and Peria; 2003; Ghosh et al., 2003; Rogoff et al., 2004; Angkinand and Willett, 2011) and currency crisis (Bubula and Ötker-Robe, 2003; Ghosh et al., 2003; Rogoff et al., 2004) in countries with less flexible exchange rate regimes. Our own empirics (Table 3, col. [6]) suggest that less flexible exchange rate arrangements are indeed associated with more banking crises, but that the relationship is not monotonic. Intermediate exchange rate regimes are about twice as likely to experience a banking crisis as a hard peg and about four times as likely as a float. Delving deeper into the intermediate regimes, it is crawling arrangements and horizontal bands that are the most crisis-prone (with about 7 percent of them experiencing a banking crisis), followed by basket pegs; while managed floats are the least likely to experience a banking crisis—and no more likely than pure floats.

Results from the probit model confirm these casual observations and show that intermediate exchange rate regimes are significantly more likely to experience banking crises than pure floats (Table 10, col. [1]). Among intermediate regimes, basket pegs, horizontal bands, and crawling pegs have a significantly greater banking crisis propensity; the coefficients on single currency pegs and managed floats are statistically insignificant (col. [2]).

Table 10.

Banking and Currency Crises: IMF’s De Facto Classification, 1980-2011

article image
Note: Dependent variable is a binary variable with one indicating a banking crisis in cols. [1]-[3] and a currency crisis in cols. [4]-[6], and zero otherwise. See Appendix B for variable definitions and datasources. All regressors are lagged one period. All specifications are estimated using the probit model. Constant included in all specifications. Statistics in parentheses reflect standard errors clustered at the country level. ***, **, and * indicate statistical significance at the 1,5 and 10 percent levels, respectively.

If we add other variables that have been identified as important determinants of banking crisis in earlier literature (e.g., Demirgüç-Kunt and Detragiache, 1998; Angkinand and Willet, 2011) such as real exchange rate overvaluation, banking system foreign liabilities, domestic credit expansion, and net capital flows (in percent of GDP), we find the estimated coefficients of these variables to be statistically significant and of the expected signs. Thus, an increase in overvaluation, faster domestic credit expansion, excessive foreign borrowing, and larger inflows are associated with a higher banking crisis probability.22 Since these variables may themselves be influenced by the regime (Tables 4-7), the addition of these explanatory variables can have three possible effects on the magnitude of the estimated coefficients of the regime variables in (2): leave them unchanged, decrease them, or increase them. To the extent that the regime coefficients remain unchanged, it means that the greater crisis propensity of some regimes is unrelated to these vulnerabilities. If the coefficient declines (a fortiori, becomes insignificant), then the crisis susceptibility of the regime is through these channels only (if it turns negative, then the regime is less susceptible to crisis than would be expected on the basis of how it scores on these vulnerabilities); and if it increases, then the regime is more susceptible to crisis than its vulnerabilities would imply.

The estimated coefficients on basket pegs, horizontal bands, and crawling pegs diminish in magnitude, and become statistically insignificant in the case of basket pegs, with the inclusion of the additional variables (Table 10, col. [3]). The result is not surprising since, as discussed above, these regimes tend to have the largest financial-stability and macroeconomic vulnerabilities; controlling for them, the regimes become less important. More surprising is that hard pegs score significantly worse on most of these risks (Table 3, cols. [1]-[5]), yet suffer fewer banking crises. One reason may be that, knowing the strictures imposed by the hard peg, including on LOLR operations (Angkinand and Willet, 2011), banking supervision is tighter, and other compensatory mechanisms are built into the design of the regime.23 Another reason may be that around one-third of banking crises are preceded by currency crises (Table 9) and, hard pegs tend to have fewer of these, as shown in Table 3.

Looking at currency crisis, these are almost five times as likely under an intermediate regime than under a hard peg, and twice as likely as under a pure float (Table 3, col. [7])—though the differences are not statistically significant (Table 10, col. [5]). Within intermediate exchange rate regimes, only crawling pegs exhibit a statistically significantly higher frequency of crisis than pure floats (col. [6]). Controlling for real exchange rate overvaluation, banking system foreign liabilities, the current account balance, and foreign exchange reserves (all of which are statistically significant with the expected signs), the coefficient on crawling pegs becomes insignificant, while the coefficient on hard pegs becomes negative and statistically significant at the 10 percent level (col. [6]). In other words, hard pegs have fewer currency crises than would be expected given their macroeconomic and financial vulnerabilities. Presumably, the greater policy discipline imposed by the hard peg together with the reluctance of speculators to take on a central bank committed to defending the parity, enables hard pegs to handle these risks without experiencing a currency crisis.

Sovereign debt crises and growth collapses

Unconditionally, the likelihood of a sovereign debt crisis is the same under hard pegs and intermediate exchange rate regimes (2 percent of the observations)—and around four times as large as under a pure float (Table 3, col. [8]). Among the intermediate exchange rate regimes, single currency and crawling pegs exhibit the highest crisis probability, roughly twice that under managed floats. None of these differences are statistically significant, however, regardless of whether other control variables (real exchange rate overvaluation, reserves, fiscal balance, real GDP growth, inflation—each of which is statistically significant with the expected sign) are included in the model (Table 11, cols. [1]-[3]).24

Table 11.

Sovereign Debt Crisis and Growth Collapses: IMF’s De Facto Classification, 1980-2011

article image
Note: Dependent variable is a binary variable with one indicating a soverein debt crisis in cols. [1]-[3] and growth collapse in cols.[4]-[6], and zero otherwise. See Appendix B for variable definitions and data sources. All regressors (except for trading partner growth) are lagged one period. All specifications are estimated using the probit model. Constant included in all specifications. Statistics in parentheses reflect standard errors clustered at the country level. ***, **, and * indicate statistical significance at the 1, 5 and 10 percent levels, respectively.