Journal Issue
Share
Article

Exits from Heavily Managed Exchange Rate Regimes

Author(s):
Enrica Detragiache, Eisuke Okada, and Ashoka Mody
Published Date:
February 2005
Share
  • ShareShare
Show Summary Details

I. Introduction

As more and more countries, especially emerging markets and developing economies, abandon tightly managed exchange rate regimes in favor of more flexibility, the question of when and how to effect the transition is widely debated. Recently, the debate has been particularly intense in reference to China, whose authorities have voiced their intention of moving away from the U.S. dollar peg, but have not acted so far.2

A widely accepted advice has been to exit when the going is good. Exit is best undertaken when the exchange rate is not under speculative pressure to depreciate; better still, some would argue, the exit should occur when the exchange rate is likely to strengthen (Eichengreen and Masson, 1998). In his characteristically lucid manner, Eichengreen (2004, page 5) explains:

There is never a convenient time to abandon a currency peg following an exchange-rate-based stabilization. But the easiest time to do so is when capital is flowing in and the exchange rate is strong. If the authorities wait too long, capital flows may have turned around in response to a deceleration in growth, problems in the banking system, or another negative event. At that point, having a more flexible exchange rate will be essential. But obtaining it smoothly, in the face of adverse speculation and without further disturbing already volatile expectations, will be nigh well impossible.

The reason for avoiding exit under pressure to depreciate is that national authorities may lose control, confidence in the country’s prospects may weaken, and the costs may be borne in the form of a heavy output loss (typically lasting between one and two years).

In practice, judging whether conditions are right to change the exchange rate regime and choosing what alternative regime to adopt is likely to be tricky, and the change in regime may bring about speculative pressures. Given the risks involved, policymakers may prefer to keep the status quo as long as the times are good (Agénor, 2004). In particular, moving from a peg to a band may create the expectation that the band is likely to be widened in the future, inviting speculators to test the government’s resolve to maintain the band. Equally, when the reason to introduce flexibility is to allow for currency depreciation to counter the overvaluation of the real rate, the determination of the extent of overvaluation is typically not straightforward. Frankel (1999) raises a more serious question of the viability of original strategy of the decision to peg or manage the exchange rate when the peg is deployed to break persistent high inflation. If this peg is to be followed by a flexible regime, knowing that a future depreciation is likely, he asks: “will the stabilization be credible in the present?” Thus, writing in September 1999, when by the standards of the second half of the 1990s conditions were calm, Frankel concluded: “Argentina seems to have done well, all things considered, by sticking with a binding commitment” (pp. 27).

Our purpose in this paper is a simple one: to characterize the types of exits from a heavily managed to a more flexible regime that have occurred since 1980. Have countries followed the approach of exiting when economic conditions are favorable? If they have not, has the outcome always been a disorderly exit with high costs? And, to the extent we observe both orderly and disorderly exits, are there identifiable differences in the conditions under which these two forms of exit occur?

We use the Reinhart and Rogoff (2004) system to classify countries into exchange rate regimes. This system has several advantages relative to other classification systems. For instance, it captures the actual behavior of monetary authorities and the foreign exchange market rather than the official exchange rate classification reported to the IMF. It provides a very detailed breakdown of observed regimes and it takes into account the presence of active parallel markets.3 Obviously, an accurate regime definition is essential to identify and date transitions to more flexible arrangements. For our purposes, an additional advantage of the Reinhart-Rogoff classification is that it offers a natural definition of orderly and disorderly exits, since it identifies a “freely falling” category, one in which the country experiences a high rate of inflation and/or a speculative attack and large depreciation of the currency. We take the transition to the freely falling category to be a “disorderly” one, while all other transitions to more flexible regimes are considered orderly.

We have three main findings. First, a simple tabulation of the data shows that in the past 20 years or so, the vast majority of exits were followed by a depreciation of the nominal exchange rate. Second, in about half the episodes, the exit was orderly and did not lead to a currency crisis or high inflation within the following 12 months. Third, based on a multinomial analysis that distinguishes between no exit, orderly exit, and disorderly exit, we find no robust differences between the general circumstances of orderly and disorderly exits. On the other hand, exits (of either type) do differ from tranquil times, in that the real exchange rate is more overvalued, the country loses reserves, and the government steps up its borrowing from the central bank. In addition, we find exits to be more likely in periods of high international interest rates and when pegs are not well established yet.

These findings indicate that, in practice, countries do not heed the advice to move away from heavily managed exchange rate regimes when the going is good, but rather wait until the parity is under pressure to depreciate. Nonetheless, the outcome is not always disastrous, as about half of the time a crisis is averted. Unfortunately, the data does not offer clear indications as to what circumstances best improve the chances of an orderly transition to greater exchange rate flexibility.

The rest of this paper is organized as follows. In Section II, we present data on episodes in which more exchange rate flexibility was introduced. Section III describes the empirical methodology and the explanatory variables. Section IV presents the results of a multinomial and binomial logit analysis. Section V contains a review of related research. The final section presents some concluding remarks.

II. The Frequency and Features of Orderly and Disorderly Exits

The Reinhart-Rogoff “natural” classification categorizes exchange rate regimes into 6 coarse and 15 fine groups (Table 1). We define an exit as a move to a more flexible exchange rate regime. More specifically, an exit occurs when a country moves from coarse categories 1–2, corresponding to pegs or heavily managed exchange rate regimes, to coarse categories 3–6, corresponding to more flexible regimes. Furthermore, a disorderly exit is one in which the transition is to the “freely falling” category, either immediately or within 12 months of the original exit. According to Reinhart and Rogoff (2004), the exchange rate is freely falling if its rate of depreciation is large, there is high inflation, or a speculative attack against the currency takes place.4 In the empirical work, we will examine the sensitivity of the results to alternative definitions of exit.

Table 1.Reinhart-Rogoff Natural Exchange Rate Classification
CoarseFineRegime
11No separate legal tender
12Preannounced peg
13Preannounced horizontal band
14De facto peg
25Preannounced crawling peg
26Preannounced crawling band (narrow)
27De facto crawling peg
28De facto crawling band (narrow)
39Preannounced crawling band (wide)
310De facto crawling band (wide)
311Moving band
312Managed floating
413Freely floating
514Freely falling
615Dual market/no parallel market data

In the period 1980-2001, there were 156 periods of contiguous observations with a heavily managed exchange rate regime (categories 1–2), often more than one per country. Of the total number of spells, 63, or 40 percent, ended in an exit. Flexibility was introduced in an orderly manner in 32 instances while the remaining 31 were disorderly exits (Table 2).5 Thus, in contrast with the earlier findings of Eichengreen et al. (1998), our sample based on the Reinhart-Rogoff regime classification indicates that orderly exits are possible and just as common as disorderly ones.6

Table 2.Exits to More Flexible Exchange Rate Regimes, 1980–2001
CountryYearMonthFromToLength

(months)
6-Month

Depreciation

(in percent)
12-Month

Depreciation

(in percent)
Anchor
Orderly exits
Australia198211De facto crawling band (narrow)Managed floating5156.815.0USD
Burundi19859De facto crawling band (narrow)De facto crawling band (wide)549-7.1-9.5USD
China19813De facto crawling band (narrow)Managed floating8712.615.1USD
Colombia198310De facto crawling band (narrow)Managed floating11513.027.3USD
Czech Republic19963De facto crawling band (narrow)De facto crawling band (wide)67-1.9-4.2DM
El Salvador19828De facto crawling band (narrow)Managed floating5120.00.0USD
Greece19817De facto crawling band (narrow)Managed floating37510.224.7USD
Guinea20005De facto crawling band (narrow)Managed floating1683.419.9USD
Haiti19895De facto crawling band (narrow)De facto crawling band (wide)5930.00.0USD
Honduras19854Pre announced pegDe facto crawling band (wide)4180.00.0USD
Hungary19991De facto crawling band (narrow)Pre announced crawling band (wide)560.65.8DM
Iceland200010De facto crawling band (narrow)Managed floating1707.813.9DM
Iraq19821Pre announced pegManaged floating5050.01.1USD
Israel19891De facto crawling band (narrow)Pre announced crawling band (wide)2513.819.9USD
Israel19912De facto crawling band (narrow)De facto crawling band (wide)1211.513.8USD
Jamaica19935De facto pegDe facto crawling band (wide)512.627.6USD
Kenya19871Pre announced pegManaged floating5656.011.8SDR
Madagascar19857De facto crawling band (narrow)Managed floating14410.019.5FRC
Mauritania198311De facto crawling band (narrow)De facto crawling band (wide)5274.614.2USD
Mauritius19826De facto crawling band (narrow)De facto crawling band (wide)785.011.6USD
Nepal19923De facto crawling band (narrow)De facto crawling band (wide)1260.09.8USD
New Zealand19853De facto crawling band (narrow)Managed floating543-15.6-14.3AUS
Paraguay19997De factor crawling pegDe facto crawling band (wide)10213.117.8USD
Philippines19935De facto crawling band (narrow)De facto crawling band (wide)999.79.8USD
Singapore199812De facto crawling band (narrow)Managed floating7080.21.2USD
Slovak Republic19979De facto crawling band (narrow)De facto crawling band (wide)540.1-2.0DM
Sri Lanka20001Pre announced crawling band (narrow)Pre announced crawling band (wide)3313.89.0USD
Sweden199212De facto crawling band (narrow)Managed floating49022.826.9DM
United Kingdom19929Pre announced horizontal bandManaged floating2416.616.2DM
Venezuela19833Pre announced pegManaged floating5190.11.3USD
Zimbabwe19837De facto crawling band (narrow)Managed floating418.925.1USD
Disorderly exits
Argentina19813Pre announced crawling pegFreely falling2794.3198.3USD
Argentina19864Pre announced pegFreely falling1113.747.9USD
Argentina200112Pre announced pegFreely falling129114.8186.2USD
Brazil19869Pre announced pegFreely falling78.7109.5USD
Brazil19894Pre announced pegFreely falling4146.21749.0USD
Brazil19992Pre announced crawling band (narrow)Freely falling5644.554.3USD
Chile19826Pre announced pegFreely falling5346.868.7USD
Costa Rica198010Pre announced pegManaged floating7838.478.9USD
Ecuador19823Pre announced pegFreely falling10918.525.5USD
Ecuador199710De facto crawling band (narrow)Freely falling811.128.9USD
Finland19929De facto crawling band (narrow)Freely falling30019.826.1DM
Guatemala198412Pre announced peg→*Dual market/no parallel market data2590.00.0USD
Guatemala19896De factor crawling pegFreely falling124.323.1USD
Indonesia19978De factor crawling pegFreely falling22588.0217.6USD
Israel19869Pre announced crawling band (narrow)Freely falling122.45.3USD
Italy19929De facto crawling band (narrow)Freely falling11717.120.9DM
Jamaica199010Pre announced pegFreely falling13815.637.8USD
Jordan198810Pre announced pegFreely falling58635.146.4SDR
Korea199712De factor crawling pegFreely falling28464.358.7USD
Laos19971De facto crawling band (narrow)Freely falling809.436.8USD
Malawi19978Pre announced pegFreely falling3219.541.6USD
Malaysia19978De facto crawling band (narrow)Freely falling69236.945.8USD
Mexico19822De factor crawling pegFreely falling6073.1151.6USD
Mexico19942De facto peg→*Pre announced crawling band (wide)635.914.8USD
Moldova19986De facto pegFreely falling4014.756.8USD
Philippines19977De facto pegFreely falling2323.738.1USD
Poland19916Pre announced pegFreely falling1816.924.1USD
Tajikistan199810Pre announced pegFreely falling1222.545.0USD
Thailand19977De facto pegFreely falling49942.457.1USD
Uganda198910Pre announced pegFreely falling3873.9109.7USD
Uruguay198212Pre announced crawling pegFreely falling50137.0166.3USD
Uruguay199112Pre announced crawling band (narrow)Freely falling1324.451.4USD
Note: The two exits with asterisks were followed by “freely falling” within the subsequent twelve months and thus considered to be disorderly.There are some episodes whose depreciation rates appear zero, but these are in fact the ones in which multiple exchange rates existed, and the market rates were depreciating even though the official rates (shown above) were not.USD, DM, FRC, AUS, SDR stand for U.S. dollar, Deutsche mark, French franc, Australian dollar, and SDR, respectively.Source: Reinhart and Rogoff (2004), International Financial Statistics, and authors’ calculations.
Note: The two exits with asterisks were followed by “freely falling” within the subsequent twelve months and thus considered to be disorderly.There are some episodes whose depreciation rates appear zero, but these are in fact the ones in which multiple exchange rates existed, and the market rates were depreciating even though the official rates (shown above) were not.USD, DM, FRC, AUS, SDR stand for U.S. dollar, Deutsche mark, French franc, Australian dollar, and SDR, respectively.Source: Reinhart and Rogoff (2004), International Financial Statistics, and authors’ calculations.

The average duration of a spell was 199 months, and spells ending in disorderly exits were considerably shorter than those ending in orderly ones (Figure 1). Most of the exits occurred in non-emerging developing countries, but this is just a reflection of the larger number of such countries in the sample (Figure 2).7 In relative terms, exits were more frequent among emerging countries. The highest concentration was that of disorderly exits in emerging markets during the 1980s, corresponding to the international debt crisis.

Figure 1.Exits by Duration

Source: Reinhart and Rogoff (2004) and authors’ calculations.

Figure 2.Exits by Country Group and Period

Note: percentages are sample frequencies measured as the number of exits divided by the total observations of spells in the country group during the period.

Source: Reinhart and Rogoff (2004) and authors’ calculations.

Next, to get a sense for whether exits occurred when the parity was under pressure to appreciate or depreciate, we turn to the behavior of the nominal exchange rate following the change in regime. Column 7 in Table 2 shows the rate of change of the exchange rate vis-à-vis the reference currency (usually the U.S. dollar) in the six months following the exit relative to the previous six months. Except for three cases, exits, including orderly ones, were followed by a depreciation of the nominal exchange rate. One can see similar results when a 12-month window is used. This suggests that, contrary to the policy recommendation of the conventional wisdom, countries facing pressures to let the exchange rate appreciate rarely respond by making the exchange rate regime more flexible. It is only when there are pressures to devalue that policymakers go for regime change.

What might explain this asymmetry is clear: when the exchange rate is weak, the country will run out of reserves unless action is taken, while when the exchange rate is strong there is no identifiable upper bound to how much foreign exchange can be accumulated. Nonetheless, the near absence of cases in which a movement toward flexibility is followed by an appreciation is perhaps surprising. Of course, this behavior may just reflect the myopia of policymakers or other distortions, and need not be optimal.8

Next, we turn to an econometric model to conduct a more rigorous investigation of the circumstances that lead countries to introduce more exchange rate flexibility and determine the orderly or disorderly character of the transition.

III. Determinants of Exits

A. Methodology and Data

To study the circumstances surrounding the introduction of more flexibility in the exchange rate regime, we estimate a multinomial logit econometric model, which distinguishes orderly and disorderly exits from “tranquil” times. In this model, the probability of exiting a heavily managed regime in an orderly or disorderly fashion relatively to the probability of not exiting is estimated as a function of several explanatory variables. More formally, let the letters t, o, and d denote the three possible events (tranquil time, disorderly exit, and orderly exit), let β be a vector of coefficients to be estimated, and let X a vector of explanatory variables. Choosing tranquil times as the base category, the multinomial logit model can be written as

for e = d, o. So for each outcome (orderly exit and disorderly exit) the coefficients to be estimated (the β’s) represent the effect of a change in the explanatory variable on the logarithm of the ratio of the probability of that outcome to the probability of a tranquil observation.

The model is estimated on a panel of observations containing all the spells during which the exchange rate was tightly managed in the Reinhart-Rogoff sample over 1980-2001. Observations are classified as exits if they refer to the month before more flexibility is introduced. In alternative specifications, observations are classified as exits for the entire six or twelve month periods before a transition. The wider window takes into account that the changes in the explanatory variables that trigger the change in regime may occur some time before the actual transition. Comparing the three alternative specifications also allows us to assess if the factors associated with the exits were exercising their influence at these different time horizons. Exits are considered disorderly if the new regime is coded as freely falling either right away or within the twelve month following an exit. Exits are classified as orderly otherwise.

The multinomial logit also allows us to formally test the hypothesis that orderly and disorderly exits are indistinguishable events with the respects to the independent variables. This is done through a Wald test. Should this hypothesis not be rejected, then the data would indicate that the appropriate model is a bivariate logit, which only discriminates between tranquil times and exits.

Although exit episodes in the Reinhart-Rogoff datasets number over 60, data limitations constrain our econometric exercise to 40 episodes in the benchmark specification, of which 18 are orderly and 22 disorderly.9 Thus, although the total number of observations is large (about 9,500) the sample is still quite small because exits are rare events. Estimation is by maximum likelihood. The standard errors are clustered by country to allow for possible correlation of the error term within each country. We find clustered standard errors to be markedly larger than robust standard errors in these data, suggesting that failure to cluster may lead to over rejections of the null hypothesis of no significant effect.

B. The Explanatory Variables

A number of factors identified in the exchange rate regime literature have been included as explanatory variables. The deviation of the real exchange rate from a moving average of the previous five years captures a possible misalignment in the real exchange rate which may contribute to the imbalance of the external accounts. Changes in foreign exchange reserves indicate pressures on the parity. Real economic conditions and trade performance are captured by export growth. Government borrowing from the central bank is introduced to test whether exits are triggered by potential inconsistencies between fiscal and exchange rate policy, as emphasized by first generation models of balance of payments crises. Private credit growth may signal a credit boom ushering in financial sector vulnerabilities that destabilize the exchange rate regime, as in the Asian crises. Trade openness may also affect the size of external shocks and the ability of an economy to respond to such shocks under limited exchange rate flexibility. GDP per capita controls for the level of development of the country, and the U.S. real interest rate captures global macroeconomic conditions. Finally, the logarithm of the number of months since the peg began measures the duration of the exchange rate regime. Duration may affect the credibility of the regime, or it may proxy unobserved country characteristics that affect the likelihood of exit. Details on the construction, sources, and summary statistics for the explanatory variables are in the Appendix.

Table 3 shows the means and standard deviations of the explanatory variables for the three categories of observations, tranquil times, orderly, and disorderly exits. Some differences across means are apparent, though standard deviations are quite large: for instance, before disorderly exits the real exchange rate is more overvalued than in tranquil times, private credit growth is faster, government borrowing from the central bank accelerates, reserves decline, and export grow is slower. Differences between orderly exits and tranquil times are typically less pronounced, but mostly go in the same direction, with the exception of private credit growth, which is slower before disorderly exits than before orderly ones. Thus, differences in means point in the direction of exits—of either type—being preceded by deteriorating economic conditions. In addition, exit observations tend to be “younger” in terms of the age of the peg than tranquil ones, the more so for disorderly exits, indicating that less well-established regimes may be more prone to change.

Table 3.Summary Statistics of Explanatory Variables
Six-Month Window
Obs.MeanS.D.Min.Max.
Real exchange rate appreciationTranquil9528-0.0390.150-0.5840.557
Orderly105-0.0290.139-0.3290.263
Disorderly1240.0250.100-0.2520.228
Trade opennessTranquil95280.4880.2420.0001.587
Orderly1050.4130.1890.0861.136
Disorderly1240.4360.2840.1201.556
Private credit growthTranquil95280.0740.146-0.9750.907
Orderly1050.0500.093-0.1880.236
Disorderly1240.1120.122-0.1580.608
Government borrowing growthTranquil95280.0550.467-1.1117.746
Orderly1050.3640.817-0.5034.332
Disorderly1240.1820.731-0.8712.867
Changes in scaled reservesTranquil95280.0050.314-1.5291.520
Orderly105-0.0350.320-1.3600.788
Disorderly124-0.0650.428-1.4921.515
Export growthTranquil95280.0510.294-1.0002.011
Orderly105-0.0470.270-0.7890.683
Disorderly1240.0170.196-0.4550.803
Real GDP per capitaTranquil95287.91411.1710.15146.895
Orderly1054.1156.4230.19426.936
Disorderly1245.0786.0780.16424.355
U.S. money market rateTranquil95280.0300.019-0.0420.087
Orderly1050.0370.021-0.0030.087
Disorderly1240.0290.022-0.0420.079
Duration of spellsTranquil9528294.157233.6951744
Orderly105271.143215.7342593
Disorderly124176.903194.0253692
Source: Reinhart and Rogoff (2004), International Financial Statistics, World Development Indicators, and authors’ calculations.
Source: Reinhart and Rogoff (2004), International Financial Statistics, World Development Indicators, and authors’ calculations.

C. Results from the Multinomial Logit

The first three columns in Table 4 present the determinants of the probability of orderly and disorderly exits relative to tranquil times for three different windows (1 month, 6 months, and 12 months before the exit). The fourth and fifth column contain variants of the benchmark model using a more restrictive and a less restrictive definition of heavily managed exchange rate regime (categories 1-4 and categories 1-11, while the benchmark is categories 1-8), and correspondingly different definitions of exit.

Table 4.Multinomial Logit Estimation Results (Tranquil as Base Category)
Benchmark SpecificationStricter PegLooser Peg
1 Month6 Month12 Month6 Month6 Month
Orderly exits
Appreciation of real exchange rate (t-1)0.61

[0.37]
1.31

[0.87]
1.69

[1.18]
2.96

[3.55]***
1.89

[1.43]
Trade openness (t-1)-0.21

[0.21]
-0.83

[0.96]
-0.74

[0.84]
-1.78

[1.06]
-1.63

[0.89]
Annual private credit growth (t-1)-0.33

[0.31]
-1.45

[1.29]
-1.88

[1.76]*
-0.01

[0.01]
-2.94

[2.83]***
Annual govn’t borrowing growth (t-1)0.51

[3.35]***
0.49

[3.01]***
0.48

[2.64]***
0.74

[0.72]
0.55

[2.22]**
Changes in scaled reserves (t-1)-0.29

[0.47]
-0.23

[0.74]
-0.02

[0.07]
-0.51

[1.58]
-0.20

[0.67]
Annual real export growth (t-1)-0.98

[1.75]*
-1.01

[1.70]*
-0.63

[1.47]
-0.16

[0.37]
-0.62

[0.97]
Real GDP per capita (USD) (t-1)-0.06

[1.69]*
-0.04

[1.40]
-0.03

[1.23]
-0.61

[1.25]
0.00

[0.04]
U.S. real money market rate (t-1)24.51

[1.85]*
20.39

[1.72]*
18.92

[1.72]*
44.80

[3.79]***
42.79

[4.04]***
Duration of spells-0.14

[0.69]
-0.12

[0.53]
-0.10

[0.41]
-0.71

[4.85]***
-0.06

[0.35]
Constant-6.04

[4.92]***
-3.92

[3.10]***
-3.31

[2.50]**
-0.40

[0.42]
-5.11

[4.92]***
Disorderly exits
Appreciation of real exchange rate (t-1)2.19

[1.67]*
2.69

[2.69]***
2.73

[2.75]***
2.97

[3.38]***
1.49

[1.45]
Trade openness (t-1)-0.19

[0.16]
-0.61

[0.52]
-0.45

[0.39]
-0.52

[0.45]
-1.09

[1.02]
Annual private credit growth (t-1)-0.06

[0.07]
0.49

[0.57]
0.87

[0.93]
0.56

[0.57]
0.63

[0.64]
Annual govn’t borrowing growth (t-1)0.32

[2.16]**
0.26

[1.30]
0.25

[1.06]
-0.94

[0.71]
0.40

[2.14]**
Changes in scaled reserves (t-1)-2.48

[4.01]***
-0.55

[1.68]*
-0.23

[1.21]
-0.09

[0.41]
-0.72

[2.65]***
Annual real export growth (t-1)-0.17

[0.38]
-0.51

[1.45]
-0.49

[1.34]
-0.24

[0.50]
-0.36

[0.87]
Real GDP per capita (USD) (t-1)-0.02

[1.09]
-0.03

[1.41]
-0.03

[1.39]
-0.14

[3.58]***
-0.03

[1.53]
U.S. real money market rate (t-1)7.01

[0.54]
3.28

[0.25]
6.75

[0.57]
16.69

[0.84]
9.36

[0.84]
Duration of spells-0.30

[2.27]**
-0.30

[2.13]**
-0.31

[2.09]**
-0.51

[3.39]***
-0.19

[1.45]
Constant-5.00

[4.51]***
-2.58

[2.55]**
-2.04

[2.04]**
-1.54

[1.73]*
-3.07

[3.81]***
Observations975797579757438611437
p-value of a Wald test0.0230.3630.2080.5480.008
(H0: Orderly and disorderly exits are indistinguishable)
Source: Reinhart and Rogoff (2004), International Financial Statistics, World Development Indicators, and authors’ calculations.Note: * significant at 10 percent; ** significant at 5 percent; *** significant at 1 percent.
Source: Reinhart and Rogoff (2004), International Financial Statistics, World Development Indicators, and authors’ calculations.Note: * significant at 10 percent; ** significant at 5 percent; *** significant at 1 percent.

A number of factors distinguish exits of either type from tranquil periods: first, exits, particularly disorderly ones, are preceded by an overvalued real exchange rate. Not surprisingly, the effect is particularly in evidence in the variant using the strictest definition of managed exchange rate, when the nominal exchange rate has hardly any flexibility. Losses in reserves are significant for disorderly exits, though not in all specifications, while an acceleration in government borrowing from the central bank and higher U.S. interest rates seem to precede orderly exits. Regimes ending in a disorderly exit are more likely to be short-lived.

Although there are some differences among the determinants of the two types of exit, a Wald test of whether the two events are indistinguishable rejects the null only when the window is one month and in the variant in which a looser definition of managed exchange rate is considered.

Another way to gauge the difference between orderly and disorderly exits is to reestimate the multinomial logit using orderly exits as the base category. The coefficients of the probability of disorderly exits, then, indicate which variables increases the probability of a disorderly exit relative to that of an orderly exit (Table 5). While some variables are significant in some specifications, no explanatory variable is robust. So the multinomial logit suggests that, at least with regard to the explanatory variables considered here, there is no significant robust difference between the circumstances preceding orderly and disorderly exits.

Table 5.Multinomial Logit Estimation Results (Orderly as Base Category)
Benchmark SpecificationStricter PegLooser Peg
1 Month6 Month12 Month6 Month6 Month
Tranquil times
Appreciation of real exchange rate (t-1)-0.61

[0.37]
-1.31

[0.87]
-1.69

[1.18]
-2.96

[3.55]***
-1.89

[1.43]
Trade openness (t-1)0.21

[0.21]
0.83

[0.96]
0.74

[0.84]
1.78

[1.06]
1.63

[0.89]
Annual private credit growth (t-1)0.33

[0.31]
1.45

[1.29]
1.88

[1.76]*
0.01

[0.01]
2.94

[2.83]***
Annual govn’t borrowing growth (t-1)-0.51

[3.35]***
-0.49

[3.01]***
-0.48

[2.64]***
-0.74

[0.72]
-0.55

[2.22]**
Changes in scaled reserves (t-1)0.29

[0.47]
0.23

[0.74]
0.02

[0.07]
0.51

[1.58]
0.20

[0.67]
Annual real export growth (t-1)0.98

[1.75]*
1.01

[1.70]*
0.63

[1.47]
0.16

[0.37]
0.62

[0.97]
Real GDP per capita (USD) (t-1)0.06

[1.69]*
0.04

[1.40]
0.03

[1.23]
0.61

[1.25]
0.00

[0.04]
U.S. real money market rate (t-1)-24.51

[1.85]*
-20.39

[1.72]*
-18.92

[1.72]*
-44.80

[3.79]***
-42.79

[4.04]***
Duration of spells0.14

[0.69]
0.12

[0.53]
0.10

[0.41]
0.71

[4.85]***
0.06

[0.35]
Constant6.04

[4.92]***
3.92

[3.10]***
3.31

[2.50]**
0.40

[0.42]
5.11

[4.92]***
Disorderly exits
Appreciation of real exchange rate (t-1)1.58

[0.73]
1.39

[0.76]
1.04

[0.61]
0.02

[0.01]
-0.41

[0.25]
Trade openness (t-1)0.02

[0.01]
0.22

[0.14]
0.28

[0.19]
1.26

[0.67]
0.53

[0.26]
Annual private credit growth (t-1)0.27

[0.20]
1.94

[1.46]
2.76

[2.12]**
0.56

[0.36]
3.56

[2.40]**
Annual govn’t borrowing growth (t-1)-0.19

[0.97]
-0.23

[1.05]
-0.22

[0.90]
-1.68

[1.04]
-0.15

[0.54]
Changes in scaled reserves (t-1)-2.19

[2.53]**
-0.32

[0.71]
-0.21

[0.65]
0.42

[1.15]
-0.52

[1.25]
Annual real export growth (t-1)0.81

[1.16]
0.51

[0.75]
0.14

[0.25]
-0.08

[0.13]
0.26

[0.39]
Real GDP per capita (USD) (t-1)0.04

[1.10]
0.01

[0.38]
0.01

[0.20]
0.47

[0.94]
-0.03

[0.80]
U.S. real money market rate (t-1)-17.50

[0.96]
-17.12

[1.00]
-12.17

[0.78]
-28.11

[1.20]
-33.43

[2.21]**
Duration of spells-0.15

[0.71]
-0.18

[0.72]
-0.21

[0.74]
0.20

[1.07]
-0.13

[0.63]
Constant1.04

[0.64]
1.34

[0.82]
1.27

[0.75]
-1.14

[0.86]
2.04

[1.55]
Observations975797579757438611437
Source: Reinhart and Rogoff (2004), International Financial Statistics, World Development Indicators, and authors’ calculationNote: * significant at 10 percent; ** significant at 5 percent; *** significant at 1 percent.
Source: Reinhart and Rogoff (2004), International Financial Statistics, World Development Indicators, and authors’ calculationNote: * significant at 10 percent; ** significant at 5 percent; *** significant at 1 percent.

We subject this conclusion to additional sensitivity tests (not reported). For instance, we include high inflation episodes (defined as observations with inflation exceeding 40 percent per year), and we control for U.S. GDP growth. Also for these alternative models the hypothesis that orderly and disorderly exits are indistinguishable cannot be rejected.

Besides the small sample size, a possible reason for lack of robust results is that the decision to exit may be non-monotonic with respect to some of the explanatory variables. Specifically, countries may be more likely to introduce flexibility both when reserves grow strongly and when reserves decline rapidly, since in both cases there are pressures on the parity. Similarly, a very overvalued or undervalued real exchange rate may prompt a move to a more flexible regime. However, a visual inspection of the frequency distribution of the various explanatory variables by category (tranquil, orderly exit, disorderly exit) does not suggest non-monotonicities of this sort. We also rerun the benchmark model splitting the change in reserves between gain and losses, and find that gains in reserves do not develop a positive and significant coefficient, suggesting that non-monotonicities are absent (Table 6).

Table 6.Multinomial Logit Results: Gains and Losses in Reserves
Benchmark SpecificationStricter PegLooser Peg
1 Month6 Month12 Month6 Month6 Month
Orderly exits
Appreciation of real exchange rate (t-1)0.62

[0.38]
1.30

[0.86]
1.69

[1.17]
2.78

[3.35]***
1.97

[1.47]
Trade openness (t-1)-0.15

[0.14]
-0.86

[0.96]
-0.74

[0.83]
-2.11

[1.28]
-1.42

[0.77]
Annual private credit growth (t-1)-0.31

[0.30]
-1.46

[1.31]
-1.89

[1.78]*
-0.11

[0.09]
-2.90

[2.83]***
Annual govn’t borrowing growth (t-1)0.51

[3.35]***
0.50

[3.01]***
0.48

[2.65]***
0.72

[0.67]
0.54

[2.16]**
Gains in scaled reserves (t-1)0.08

[0.07]
-0.41

[0.55]
-0.01

[0.02]
-2.42

[1.82]*
0.62

[0.91]
Losses in scaled reserves (t-1)0.58

[0.80]
0.08

[0.12]
0.03

[0.04]
-0.61

[0.65]
0.94

[1.70]*
Annual real export growth (t-1)-0.96

[1.74]*
-1.02

[1.69]*
-0.63

[1.47]
-0.15

[0.34]
-0.62

[1.00]
Real GDP per capita (USD) (t-1)-0.06

[1.64]
-0.04

[1.37]
-0.03

[1.20]
-0.63

[1.20]
0.00

[0.01]
U.S. real money market rate (t-1)24.79

[1.86]*
20.26

[1.73]*
18.91

[1.73]*
43.91

[3.75]***
43.67

[4.17]***
Duration of spells-0.14

[0.68]
-0.12

[0.54]
-0.10

[0.41]
-0.72

[4.87]***
-0.05

[0.30]
Constant-6.17

[5.04]***
-3.86

[3.20]***
-3.32

[2.52]**
0.13

[0.13]
-5.48

[4.76]***
Disorderly exits
Appreciation of real exchange rate (t-1)2.22

[1.69]*
2.71

[2.69]***
2.74

[2.75]***
2.88

[3.25]***
1.53

[1.50]
Trade openness (t-1)-0.05

[0.04]
-0.42

[0.37]
-0.28

[0.25]
-0.69

[0.58]
-0.97

[0.92]
Annual private credit growth (t-1)-0.04

[0.05]
0.53

[0.63]
0.91

[0.98]
0.50

[0.52]
0.64

[0.66]
Annual govn’t borrowing growth (t-1)0.32

[2.15]**
0.26

[1.30]
0.25

[1.07]
-0.93

[0.70]
0.40

[2.14]**
Gains in scaled reserves (t-1)-0.52

[0.54]
0.45

[0.96]
0.63

[1.45]
-0.88

[1.45]
-0.02

[0.04]
Losses in scaled reserves (t-1)2.77

[4.32]***
1.22

[2.22]**
0.98

[1.87]*
-0.68

[0.89]
1.13

[2.34]**
Annual real export growth (t-1)-0.16

[0.36]
-0.48

[1.43]
-0.48

[1.35]
-0.23

[0.47]
-0.36

[0.89]
Real GDP per capita (USD) (t-1)-0.02

[1.06]
-0.03

[1.34]
-0.03

[1.32]
-0.14

[3.67]***
-0.03

[1.49]
U.S. real money market rate (t-1)7.53

[0.57]
3.62

[0.28]
7.09

[0.60]
16.30

[0.83]
9.79

[0.87]
Duration of spells-0.29

[2.21]**
-0.29

[2.09]**
-0.30

[2.06]**
-0.51

[3.43]***
-0.18

[1.42]
Constant-5.29

[4.56]***
-2.92

[2.81]***
-2.37

[2.33]**
-1.28

[1.42]
-3.30

[4.10]***
Observations975797579757438611437
Source: Reinhart and Rogoff (2004), International Financial Statistics, World Development Indicators, and authors’ calculations.Note: * significant at 10 percent; ** significant at 5 percent; *** significant at 1 percent.
Source: Reinhart and Rogoff (2004), International Financial Statistics, World Development Indicators, and authors’ calculations.Note: * significant at 10 percent; ** significant at 5 percent; *** significant at 1 percent.

D. Results from the Binomial Logit Model

Since distinguishing between orderly and disorderly exits proved inconclusive, we now turn to estimating a bivariate logit model in which observations can only be exits or tranquil times. This should give us an indication of what prompts moves to more exchange rate flexibility. The results show that changes occur when there are pressures to devalue the exchange rate (Table 7): the real exchange rate is overvalued, reserves are falling, and the government is increasingly relying on the central bank for deficit financing. In addition, world interest rates tend to be higher before change in regime, suggesting that reversals in capital inflows may contribute to trigger exits. Finally, there is some evidence that less well-established regimes are more likely to be abandoned. These results are fairly robust to changing the window and the definition of exit.

Table 7.Binomial Logit Estimation Results
Benchmark SpecificationStricter PegLooser Peg
1 Month6 Month12 Month6 Month6 Month
Exits
Appreciation of real exchange rate (t-1)1.47

[1.41]
2.07

[2.38]**
2.27

[2.68]***
2.78

[4.38]***
1.59

[1.99]**
Trade openness (t-1)-0.28

[0.40]
-0.69

[0.96]
-0.55

[0.75]
-1.17

[1.08]
-1.24

[1.29]
Annual private credit growth (t-1)-0.19

[0.28]
-0.36

[0.49]
-0.29

[0.36]
0.13

[0.17]
-0.65

[0.88]
Annual govn’t borrowing growth (t-1)0.40

[3.47]***
0.38

[2.61]***
0.36

[2.11]**
-0.03

[0.04]
0.47

[2.92]***
Changes in scaled reserves (t-1)-1.66

[3.17]***
-0.40

[1.80]*
-0.14

[0.85]
-0.30

[1.43]
-0.53

[2.61]***
Annual real export growth (t-1)-0.53

[1.44]
-0.76

[2.20]**
-0.57

[1.90]*
-0.19

[0.57]
-0.47

[1.20]
Real GDP per capita (USD) (t-1)-0.04

[2.00]**
-0.03

[2.09]**
-0.03

[1.97]**
-0.21

[4.06]***
-0.02

[1.08]
U.S. real money market rate (t-1)15.35

[1.59]
11.51

[1.27]
12.37

[1.47]
30.79

[2.75]***
23.18

[2.83]***
Duration of spells-0.23

[1.79]*
-0.22

[1.72]*
-0.22

[1.63]
-0.58

[4.82]***
-0.15

[1.41]
Constant-4.66

[5.88]***
-2.44

[3.13]***
-1.84

[2.39]**
-0.48

[0.71]
-3.08

[4.96]***
Observations975797579757438611437
Total exits40229437135258
Correctly predicted exits25146265110155
Prob. of correctly predicting exits0.630.640.610.810.60
Total tranquil times971795289320425111179
Correctly predicted tranquil times66646093587930417412
Prob. of correctly predicting tranquil times0.690.640.630.720.66
Source: Reinhart and Rogoff (2004), International Financial Statistics, World Development Indicators, and authors’ calculations.Note: * significant at 10 percent; ** significant at 5 percent; *** significant at 1 percent
Source: Reinhart and Rogoff (2004), International Financial Statistics, World Development Indicators, and authors’ calculations.Note: * significant at 10 percent; ** significant at 5 percent; *** significant at 1 percent

Turning now to the performance of the model, it is customary to compare fitted probabilities with the sample frequency of each event. If the fitted probably exceeds the sample frequency, then the model provides useful information about the event. Based on the benchmark model with a six month window, the fitted probability of exit exceeds the sample frequency in 64 percent of the exit cases; the same is true for tranquil observations. The percentage of observations correctly predicted is higher (reaching 80 percent) when the definition of peg is stricter. Similar results obtain in the alternative specifications.

Table 8 contains further sensitivity tests: excluding high inflation countries does not change the results much, and neither does replacing GDP per capita with dummies for emerging and developing countries. When the real effective exchange rate is used to measure exchange rate overvaluation, this variable is not significant, although the sign remains positive, and government borrowing also loses significance. The specification in the last column contains a new explanatory variable: changes in the political regime. It appears that this variable is positively correlated with changes in the exchange rate regime, and when it is introduced the coefficients of the other explanatory variables do not change much.

Table 8.Binomial Logit Estimation Results–Robustness
Includes High

Inflation

6 Month
Real Effective

Exchange Rate

6 Month
Country Group

Dummies

6 Month
Political

Changes

6 Month
Exits
Appreciation of real exchange rate (t-1)1.72

[1.98]**
1.35

[0.95]
2.02

[2.29]**
2.17

[2.50]**
Trade openness (t-1)-1.03

[1.35]
0.12

[0.21]
-0.62

[0.90]
-0.65

[0.88]
Annual private credit growth (t-1)0.28

[0.35]
0.42

[0.53]
-0.57

[0.77]
-0.3

[0.41]
Annual govn’t borrowing growth (t-1)0.42

[3.14]***
0.07

[0.63]
0.37

[2.64]***
0.37

[2.53]**
Changes in scaled reserves (t-1)-0.31

[1.66]*
-0.45

[1.73]*
-0.39

[1.81]*
-0.4

[1.78]*
Annual real export growth (t-1)-0.73

[2.22]**
-0.45

[1.09]
-0.81

[2.25]**
-0.71

[2.18]**
Real GDP per capita (USD) (t-1)-0.03

[1.64]
-0.03

[1.59]
-0.03

[1.79]*
U.S. real money market rate (t-1)13.45

[1.57]
-1.38

[0.13]
11.83

[1.31]
12.58

[1.37]
Duration of spells-0.3

[2.71]***
-0.26

[1.86]*
-0.22

[1.68]*
-0.24

[1.85]*
MSCI emerging markets0.87

[1.72]*
Other developing countries0.6

[1.33]
Politically unstable periods0.78

[2.22]**
Constant-2.06

[3.22]***
-2.5

[3.22]***
-3.2

[3.40]***
-2.55

[3.27]***
Observations9957920897699757
Source: Reinhart and Rogoff (2004), International Financial Statistics, World Development Indicators, Polity IV Project, and authors’ calculations.Note: * significant at 10 percent; ** significant at 5 percent; *** significant at 1 percent.
Source: Reinhart and Rogoff (2004), International Financial Statistics, World Development Indicators, Polity IV Project, and authors’ calculations.Note: * significant at 10 percent; ** significant at 5 percent; *** significant at 1 percent.

IV. A Review of the Literature

Most of the empirical work on exchange rates has focused either on the choice of the regime and its economic performance or on the circumstances preceding speculative attacks.10 There is, however, a smaller literature on exits from pegs or other heavily managed regimes, whether or not they are associated with speculative attacks.

Eichengreen et al. (1998) identify changes in the exchange rate regime in developing countries using the IMF’s Annual Report of Exchange Arrangements and Exchange Restrictions (AREAER). Accordingly, recorded episodes reflect changes in the official (or de jure) regime, rather than in the de facto one. The definition of exit includes exits from single currency or basket pegs, but excludes exits from crawling pegs, target zones, and unofficially pegged regimes. 29 cases are identified during 1977-95, of which 23 are currency crises based on the definition of Frankel and Rose (1996). Hence, orderly exits are extremely rare in this sample, and the authors do not attempt to distinguish between orderly and disorderly episodes through an econometric analysis of the data. Based in part on the high rate of disorderly exits, this paper concludes that countries should introduce exchange rate flexibility in good times, when pressures are for the exchange rate to appreciate and reserves are accumulating. This view is reiterated in Eichengreen (2004).

Klein and Marion (1997) examine the duration of exchange rate pegs in Latin America using a binomial logit econometric model. In contrast with our study, in this paper, an exit need not imply a change in the exchange rate regime, but can be (and often is) simply a change in the parity. As in Eichengreen and others, the regime is identified based on the de jure IMF classification. In addition, there is no attempt to distinguish among types of exits (orderly or disorderly, new peg or more flexible regime).The main findings are that exits tend to occur when the real exchange rate is overvalued and reserves are low; trade openness and political stability are associated with more exchange regime stability; and exits becomes less likely the longer is the duration of the regime.

Using the IMF de facto classification of Bubula and Ötker-Robe (2002), Duttagupta and Ötker-Robe (2003) take a comprehensive look at changes in exchange regimes. In this study, the definition of exit includes one-off changes in the parity (including revaluations) as well as shifts to less flexible regimes.11 They further distinguish between orderly and disorderly exits, with the latter being defined as an exit accompanied by a large depreciation of the exchange rate. They find that orderly exits tend to be associated with more government borrowing and trade openness than tranquil times, while disorderly exits are associated with declining reserves, lower export revenues, and an overvalued real exchange rate.12 In contrast with Klein and Marion (1997), they find that a longer duration for the peg is more likely to trigger a crisis. Finally, at conventional significance levels the empirical model rejects the hypothesis that regime changes differ from tranquil observations in all cases, except for exits to more flexible regimes (both orderly and disorderly).

In a recent paper, Asici and Wyplosz (2003) study what sets apart orderly and disorderly exits, identified based on the Reinhart-Rogoff classification, but do not study how exits differ from “tranquil times.” The conclusions support the conventional wisdom that countries that exit when macroeconomic performance is good avoid crises. Corruption and financial depth are found to make an exit more likely to be disorderly.

IMF (2004) carries out a descriptive review of emerging market transitions toward more exchange rate flexibility using the IMF de facto classification. The focus is mostly on the evolution of monetary and financial institutions during the transition. Among the findings is that countries moving to more flexibility tend to introduce more central bank independence, move towards an inflation targeting framework, and have better bank supervision and more developed securities markets than other countries.

V. Conclusions

This paper finds that exits from managed exchange rate regimes towards more flexibility occur when the parity is under pressure to devalue, while exits when the exchange rate is under pressure to appreciate are exceedingly rare. Exits have been about evenly divided between orderly and disorderly events, but differences in economic conditions preceding orderly and disorderly exits are not sharp, and cannot be picked up by our econometric tests.

Thus, countries have not heeded the conventional policy advice to exit tight pegs when the going is good. To the contrary, they seem to have relied more on the popular wisdom summarized by the principle: “if it isn’t broken, why fix it?” Waiting until conditions deteriorate has not always proven disastrous, as exits have remained orderly in about half the cases. Nonetheless, a policymaker may wish to know what might improve chances of a smooth exit. In this respect, our findings are disappointing, because none of the variables we have studied, which capture many of the factors highlighted in the literature, helps to discriminate between orderly and disorderly exits.

Failure to uncover clear patterns may be due to the small sample size although we have data for many years and countries, exits remain relatively rare events. But our results may also point to a more fundamental indeterminacy of the effects of changes in the exchange rate regime, which in turn may explain why country authorities wait till they are left with little choice. If an exit at any time can go wrong, then postponing change is always an attractive option.

APPENDIX

DataDescription

LabelVariableSource/Definition
AReinhart-Rogoff regime classificationhttp://www.wam.umd.edu/∼creinhar/Links.html
BNominal exchange rate vis-à-vis U.S. dollarInternational Financial Statistics, line ‥RF.ZF
CConsumer price indexInternational Financial Statistics, line 64…ZF
DAnnual exportsInternational Financial Statistics, line 99C‥ZF/99C.CZF
EAnnual importsInternational Financial Statistics, line 98C‥ZF/98C.CZF
FBanks’ claims on private sectorInternational Financial Statistics, line 22D‥ZF
GCentral banks’ claims on governmentsInternational Financial Statistics, line 12A‥ZF
HTotal reserves minus goldInternational Financial Statistics, line .1L.DZF
IMoney market rateInternational Financial Statistics, line 60B‥ZF
JGross domestic productInternational Financial Statistics, line 99B‥ZF/99B.CZF
KPopulationInternational Financial Statistics, line 99Z‥ZF
LReal effective exchange rateINS/IFS/GDS
MMonthly exportsDirection of Trade Statistics, line 70‥DZD001
NMonthly importsDirection of Trade Statistics, line 71‥DZD001
OReal GDP per capitaWorld Development Indicators, line NYGDPPCAPKD
PPolity variableshttp://www.cidcm.umd.edu/inscr/polity/
QReal exchange rate appreciation vis-à-vis U.S. dollarRate of change relative to the past five-year average
RTrade openness(D + E)/J; interpolated to allow it to vary on a monthly basis
SAnnual private credit growthYear-to-year growth rate of F/C
TAnnual government borrowing growthYear-to-year growth rate of G/C
UChanges in scaled reservesH divided by the past one-year average of imports (N)
VAnnual real export growthYear-to-year growth rate of M/C
WReal GDP per capitaO interpolated to allow it to vary on a monthly basis
XU.S. real money market rateI adjusted by year-to-year U.S. inflation rate (C)
YPolitically unstable periodsSix month before and after political changes or transitional periods (P)
References

    AgénorPierre-Richard2004Orderly Exits from Adjustable Pegs and Exchange Rate BandsPolicy Issues and the Role of Capital Flowsunpublished manuscript (unpublished: Washington: The World Bank).

    • Crossref
    • Search Google Scholar
    • Export Citation

    AsiciAhmet and CharlesWyplosz2003The Art of Gracefully Exiting a PegThe Economic and Social ReviewVol. 3 (Winter) pp. 21128.

    • Search Google Scholar
    • Export Citation

    BubulaA. and InciOtker-Robe2002The Evolution of Exchange Rate Regimes Since 1990: Evidence from De Facto PoliciesIMF Working Paper No. 02/155 (Washington: International Monetary Fund).

    • Crossref
    • Search Google Scholar
    • Export Citation

    DuttaguptaRupa and InciOtker-Robe2003Exits from Pegged Regimes: An Empirical AnalysisIMF Working Paper No. 03/147 (Washington: International Monetary Fund).

    • Search Google Scholar
    • Export Citation

    EichengreenBarry2004Chinese Currency ControversiesUniversity of California. Available via the Internet: Berkeley.http://emlab.berkeley.edu/users/eichengr/research/asiancurrencies7jun9-04.pdf

    • Search Google Scholar
    • Export Citation

    EichengreenBarry and PaulMasson1998Exit Strategies: Policy Options for Countries Seeking Greater Exchange Rate FlexibilityOccasional Paper No. 98/168 (Washington: International Monetary Fund).

    • Crossref
    • Search Google Scholar
    • Export Citation

    EichengreenBarry and AndrewRose2001To Defend or Not to Defend? That Is the Question.” Available via the Internet: http://www.nber.org/∼confer/2001/ccdf/rose.pdf.

    • Search Google Scholar
    • Export Citation

    FernandezRaquel and DaniRodrik1991Resistance to Reform: Status Quo Bias in the Presence of Individual-Specific UncertaintyAmerican Economic ReviewVol. 81 pp. 114655.

    • Search Google Scholar
    • Export Citation

    FrankelJeffrey A.1999No Single Currency Regime Is Right For All Countries or at All TimesNBER Working Paper 7338 (Cambridge, Massachusetts: National Bureau of Economic Research)

    • Search Google Scholar
    • Export Citation

    HusainAasimAshokaMody and KennethRogoff2004Exchange Rate Regime Durability and Performance in Developing Countries Versus Advanced EconomiesNBER Work Paper 10673 (Cambridge, Massachusetts: National Bureau of Economic Research)

    • Search Google Scholar
    • Export Citation

    International Monetary Fund2004World Economic Outlook (Washington: International Monetary Fund).

    KleinMichael and NancyMarion1997Explaining the Duration of Exchange-Rate PegsJournal of Development EconomicsVol. 54No. 2 pp. 387404.

    • Crossref
    • Search Google Scholar
    • Export Citation

    PrasadEswarThomasRumbaugh and QingWang2005Putting the Cart Before the Horse? Capital Account Liberalization and Exchange Rate Flexibility in ChinaIMF Policy Discussion Paper No. 05/01 (Washington: International Monetary Fund).

    • Search Google Scholar
    • Export Citation

    ReinhartCarmen and KennethRogoff2004The Modern History of Exchange Rate Arrangements: A ReinterpretationQuarterly Journal of EconomicsVol. 119No. 1 pp. 148.

    • Crossref
    • Search Google Scholar
    • Export Citation
1The authors are grateful for valuable comments from Barry Eichengreen, Simon Johnson, Kalpana Kochhar, Laura Kodres, Sandy Mackenzie, Jonathan Ostry, Eswar Prasad, Raghuram Rajan, and David Robinson.
2For a comprehensive review of the policy debate on China, see Prasad and others (2005).
3For a detailed description of the methodology used and comparison with other existing classifications, see Reinhart and Rogoff (2004).
4Spells that ended with an exit to coarse category 6 (Dual market/no parallel market data) were excluded, since it was not possible to determine if the exit was orderly or disorderly. Only a handful such episodes are in the sample.
5If an orderly exit is followed by “freely falling” within the subsequent twelve months, it is classified as a disorderly exit.
6This is true also if we use a tighter definition of managed exchange rate regime, closer to that used by Eichengreen et al. (1998). Using the IMF de facto classification for 1985–2002, Duttagupta and Otker-Robe (2003) find that orderly exits were even more frequent. Specifically, they identify 41 episodes in which more flexibility was introduced in an orderly fashion and 30 episodes in which a sharp depreciation of the currency followed the exit (see Section IV below).
7As in Husain, Mody, and Rogoff (2004), emerging market countries are defined using Morgan Stanley Capital International (MSCI) classification, implying that international investors have a real interest in these economies.
8For a model explaining policymakers’ status quo bias, see for instance Fernandez and Rodrik (1991).
9As customary in cross-country studies, we exclude from the sample very small countries, defined as those with population less than one million. Also, to eliminate outliers, we exclude observations in which explanatory variables are beyond four standard deviations from the mean. Results do not change much if extreme observations are included.
10For a recent review of the first group of studies, see Rogoff and others (2004). For the latter, see among others, Eichengreen, Rose, and Wyplosz (1995) and Frankel and Rose (1996).
11The IMF de facto classification is based in part on qualitative judgment of its desk economists. Unlike the Reinhart-Rogoff classification, parallel foreign exchange markets are not taken into account.
12As in other studies, real overvaluation is measured as deviation from a linear trend estimated over the sample period. Because large nominal devaluations typically also entail a large real devaluation, which likely pulls the entire trend down, finding the real exchange rate above trend before a disorderly exit is almost tautological. Also, failure to cluster standard errors by country may lead to underestimate standard errors in this study.

Other Resources Citing This Publication