THE MODERN HISTORY OF EXCHANGE RATE ARRANGEMENTS: A REINTERPRETATION

ou juste avant la balise de fermeture -->

THE QUARTERLY JOURNAL OF ECONOMICS Vol. CXIX February 2004 Issue 1 THE MODERN HISTORY OF EXCHANGE RATE ARRANGEMENTS: A REINTERPRETATION* CARMEN M. REINHART AND KENNETH S. ROGOFF We develop a novel system of reclassifying historical exchange rate regimes. One key difference between our study and previous classi?cations is that we employ monthly data on market-determined parallel exchange rates going back to 1946 for 153 countries. Our approach differs from the IMF of?cial classi?cation (which we show to be only a little better than random); it also differs radically from all previous attempts at historical reclassi?cation. Our classi?cation points to a rethinking of economic performance under alternative exchange rate regimes. Indeed, the breakup of Bretton Woods had less impact on exchange rate regimes than is popularly believed. I. INTRODUCTION This paper rewrites the history of post-World War II exchange rate arrangements, based on an extensive new monthly data set spanning across 153 countries for 1946 –2001. Our approach differs not only from countries’ of?cially declared classi?- cations (which we show to be only a little better than random); it also differs radically from the small number of previous attempts at historical reclassi?cation. 1 * The authors wish to thank Alberto Alesina, Arminio Fraga, Amartya Lahiri, Vincent Reinhart, Andrew Rose, Miguel Savastano, participants at Harvard University’s Canada-US Economic and Monetary Integration Conference, International Monetary Fund-World Bank Joint Seminar, National Bureau of Economic Research Summer Institute, New York University, Princeton University, and three anonymous referees for useful comments and suggestions, and Kenichiro Kashiwase, Daouda Sembene, and Ioannis Tokatlidis for excellent research assistance. Data and background material to this paper are available at http://www.puaf.umd.edu.faculty/papers/reinhart/reinhart.htm. 1. The of?cial classi?cation is given in the IMF’s Annual Report on Exchange Rate Arrangements and Exchange Restrictions, which, until recently, asked member states to self-declare their arrangement as belonging to one of four categories. © 2004 by the President and Fellows of Harvard College and the Massachusetts Institute of Technology. The Quarterly Journal of Economics, February 2004 1As a ?rst innovation, we incorporate data on parallel and dual exchange rate markets, which have been enormously important not only in developing countries but in virtually all the European countries up until the late 1950s, and sometimes well beyond. We argue that any classi?cation algorithm that fails to distinguish between uni?ed rate systems (with one of?cial exchange rate and no signi?cant “black” or parallel market) and all others is fundamentally awed. Indeed, in the vast majority of multiple exchange rate or dual systems, the oating dual or parallel rate is not only a far better barometer of monetary policy than is the of?cial exchange rate, it is often the most economically meaningful rate. 2 Very frequently—roughly half the time for of?cial pegs—we ?nd that dual/parallel rates have been used as a form of “back door” oating, albeit one usually accompanied by exchange controls. The second novelty in our approach is that we develop extensive chronologies of the history of exchange arrangements and related factors, such as exchange controls and currency reforms. Together with a battery of descriptive statistics, this allows us to draw a nuanced distinction between what countries declare as their of?cial de jure regime, and their actual de facto exchange rate practices. To capture the wide range of arrangements, our approach allows for fourteen categories of exchange rate regimes, ranging from no separate legal tender or a strict peg to a dysfunctional “freely falling” or “hyperoat.” Some highlights from our reclassi?cation of exchange rate arrangements are as follows. First, dual, or multiple rates, and parallel markets have prevailed far more frequently than is commonly acknowledged. In 1950, 45 percent of the countries in our sample had dual or multiple rates; many more had thriving parallel markets. Among the industrialized economies, dual or multiple rates were the Previous studies have either extended the four-way of?cial classi?cation into a more informative taxonomy (see Ghosh et al. [1997]), or relied largely on statistical methods to regroup country practices (see Levy-Yeyati and Sturzenegger [2002]). The Fund, recognizing the limitations of its former strategy, revised and upgraded the of?cial approach toward classifying exchange rate arrangements in 1997 and again in 1999. Notably, all these prior approaches to exchange rate regime classi?cation, whether or not they accept the country’s declared regime, have been based solely on of?cial exchange rates. 2. When we refer to multiple exchange rates in this context, we are focusing on the cases where one or more of the rates is market-determined. This is very different from the cases where the multiple of?cial rates are all ?xed and simply act as a differential tax on a variety of transactions. Dual markets are typically legal, whereas parallel markets may or may not be legal. 2 QUARTERLY JOURNAL OF ECONOMICSnorm in the 1940s and the 1950s, and in some cases, these lasted until much later. Our data lend strong support to the view stressed by Bordo [1993] that Bretton Woods encompassed two very different kinds of exchange rate arrangements in the preand postconvertibility periods and that the period of meaningful exchange rate stability was quite short-lived. In the developing world, such practices remained commonplace through the 1980s and 1990s and into the present. We show that market-determined dual/parallel markets are important barometers of underlying monetary policy. This may be obvious in cases such as modern-day Myanmar where the parallel market premium at the beginning of 2003 exceeded 700 percent. As we show, however, the phenomenon is much more general, with the parallel market premium often serving as a reliable guide to the direction of future of?cial exchange rate changes. Whereas dual/parallel markets have been marginal over some episodes, they have been economically important in others, and there are many instances where only a few transactions take place at the of?cial rate. To assess the importance of secondary (legal or illegal) parallel markets, we collected data that allow us to estimate export misinvoicing practices, in many cases going back to 1948. These estimates show that leakages from the of?cial market were signi?cant in many of the episodes when there were dual or parallel markets. Second, when one uses market-determined rates in place of of?cial rates, the history of exchange rate policy begins to look very different. For example, it becomes obvious that de facto oating was common during the early years of the Bretton Woods era of “pegged” exchange rates. Conversely, many “oats” of the post-1980s turn out to be (de facto) pegs, crawling pegs, or very narrow bands. Of countries listed in the of?cial IMF classi?cation as managed oating, 53 percent turned out to have de facto pegs, crawls, or narrow bands to some anchor currency. Third, next to pegs (which account for 33 percent of the observations during 1970 –2001 (according to our new “Natural” classi?cation), the most popular exchange rate regime over modern history has been the crawling peg, which accounted for over 26 percent of the observations. During 1990 to 2001 this was the most common type of arrangement in emerging Asia and Western Hemisphere (excluding Canada and the United States), making up for about 36 and 42 percent of the observations, respectively. Fourth, our taxonomy introduces a new category: freely fallEXCHANGE RATE ARRANGEMENTS 3ing, or the cases where the twelve-month ination rate is equal to or exceeds 40 percent per annum. 3 It turns out to be a crowded category indeed, with about 12 1 2 percent of the observations in our sample occurring in the freely falling category. As a result, “freely falling” is about three times as common as “freely oating,” which accounts for only 4 1 2 percent of the total observations. (In the of?cial classi?cation, freely oating accounts for over 30 percent of observations over the past decade.) Our new freely falling classi?cation makes up 22 and 37 percent of the observations, respectively, in Africa and Western Hemisphere (excluding Canada and the United States) during 1970 –2001. In the 1990s freely falling accounted for 41 percent of the observations for the transition economies. Given the distortions associated with very high ination, any ?xed versus exible exchange rate regime comparisons that do not break out the freely falling episodes are meaningless, as we shall con?rm. There are many important reasons to seek a better approach to classifying exchange rate regimes. Certainly, one is the recognition that contemporary thinking on the costs and bene?ts of alternative exchange rate arrangements has been profoundly in- uenced by the large number of studies on the empirical differences in growth, trade, ination, business cycles, and commodity price behavior. Most have been based on the of?cial classi?cations and all on of?cial exchange rates. In light of the new evidence we collect, we conjecture that the inuential results in Baxter and Stockman [1989]—that there are no signi?cant differences in business cycles across exchange arrangements—may be due to the fact that the of?cial historical groupings of exchange rate arrangements are misleading. The paper proceeds as follows. In the next section we present evidence to establish the incidence and importance of dual or multiple exchange rate practices. In Section III we sketch our methodology for reclassifying exchange rate arrangements. Section IV addresses some of the possible critiques to our approach, compares our results with the “of?cial history,” and provides examples of how our reclassi?cation may reshape evidence on the links between exchange rate arrangements and various facets of economic activity. The ?nal section reiterates some of the main 3. We also include in the freely falling category the ?rst six months following an exchange rate crisis (see the Appendix for details), but only for those cases where the crisis marked a transition from a peg or quasi-peg to a managed or independent oat. 4 QUARTERLY JOURNAL OF ECONOMICS?ndings, while background material to this paper provides the detailed country chronologies that underpin our analysis. II. THE INCIDENCE AND IMPORTANCE OF DUAL AND MULTIPLE EXCHANGE RATE ARRANGEMENTS In this section we document the incidence of dual or parallel markets (legal or otherwise) and multiple exchange rate practices during post-World War II. We then present evidence that the market-determined exchange rate is a better indicator of the underlying monetary policy than the of?cial exchange rate. Finally, to provide a sense of the quantitative importance for economic activity of the dual or parallel market, we present estimates of “leakages” from the of?cial market. Speci?cally, we provide quantitative measures of export misinvoicing practices. We primarily use monthly data on of?cial and market-determined exchange rates for the period 1946 –2001. In some instances, the data for the market-determined rate is only available for a shorter period and the background material provides the particulars on a country-by-country basis. The pre-1999 marketdetermined exchange rate data come from various issues of Pick’s Currency Yearbook, Pick’s Black Market Yearbooks, and World Currency Reports, and the of?cial rate comes from the same sources and as well as the IMF. The quotes are end-of-month exchange rates and are not subject to revisions. For the recent period (1999 –2001) the monthly data on market-determined exchange rates come from the original country sources (i.e., the central banks), for those countries where there are active parallel markets for which data are available. 4 Since our coverage spans more than 50 years, it encompasses numerous cases of monetary reforms involving changes in the units of account, so the data were spliced accordingly to ensure continuity. II.A. On the Popularity of Dual and Multiple Exchange Rate Practices Figure I illustrates de facto and de jure nonuni?ed exchange rate regimes. The ?gure shows the incidence of exchange rate arrangements over 1950 –2001, with and without stripping out 4. These countries include Afghanistan, Angola, Argentina, Belarus, Belize, Bolivia, Burundi, Congo (DCR), Dominican Republic, Egypt, Ghana, Iran, Libya, Macedonia, Mauritania, Myanmar, Nigeria, Pakistan, Rwanda, Tajikistan, Turkmenistan, Ukraine, Uzbekistan, Yemen, Yugoslavia, and Zimbabwe. EXCHANGE RATE ARRANGEMENTS 5cases of dual markets or multiple exchange rates. The IMF classi?cation has been simpli?ed into what it was back in the days of Bretton Woods—namely, Pegs and Other. 5 The dark portions of the bars represent cases with uni?ed exchange rates, and the lightly shaded portion of each bar separates out the dual, multiple, or parallel cases. In 1950 more than half (53 percent) of all arrangements involved two or more exchange rates. Indeed, the heyday of multiple exchange rate practices and active parallel markets was 1946 –1958, before the restoration of convertibility in Europe. Note also, that according to the of?cial IMF classi?- cation, pegs reigned supreme in the early 1970s, accounting for over 90 percent of all exchange rate arrangements. In fact, over half of these “pegs” masked parallel markets that, as we shall show, often exhibited quite different behavior. 5. For a history of the evolution of the IMF’s classi?cation strategy, see the working paper version of this paper, Reinhart and Rogoff [2002]. FIGURE I The Incidence of Dual or Multiple Exchange Rate Arrangements, 1950–2001: Simpli?ed IMF Classi?cation Sources: International Monetary Fund, Annual Report on Exchange Arrangements and Exchange Restrictions and International Financial Statistics; Pick and Se´dillot [1971]; International Currency Analysis, World Currency Yearbook, various issues. Exchange rate arrangements classi?ed as “Other” include the IMF’s categories of limited exibility, managed oating, and independently oating. 6 QUARTERLY JOURNAL OF ECONOMICSII.B. The Market-Determined Exchange Rate as an Indicator of Monetary Policy While the quality of data on market-determined rates is likely to vary across countries and time, we nevertheless believe these data to be generally far better barometers of the underlying monetary policy than are of?cial exchange rates. For instance, if the laxity in monetary policy is not consistent with maintaining a ?xed of?cial exchange rate, one would expect that the marketdetermined rate starts depreciating ahead of the inevitable devaluation of the of?cial rate. When the of?cial realignment occurs—it is simply a validation of what had previously transpired in the free market. Indeed, this is the pattern shown in the three panels of Figure II for the cases of Bolivia, Indonesia, and Iran— many more such cases are displayed in the ?gures that accompany the 153 country chronologies. 6 This pattern also emerges often in the developed European economies and Japan in the years following World War II. To illustrate more rigorously that the market-based exchange rate is a better indicator of the monetary policy stance than the of?cial rate, we performed two exercises for each country. First, we examined whether the market-determined exchange rate systematically predicts realignments in the of?cial rate, as suggested in Figure II. To do so, we regressed a currency crash dummy on the parallel market premium lagged one to six months, for each of the developing countries in our sample. 7 If the market exchange rate consistently anticipates devaluations of the of?cial rate, its coef?cient should be positive and statistically signi?cant. If, in turn, the of?cial exchange rate does not validate the market rate, then the coef?cient on the lagged market exchange rate will be negative or simply not signi?cant. Table I summarizes the results of the country-by-country time series probit regressions. In the overwhelming number of cases (97 percent), the coef?cient on the market-determined exchange rate is positive. In about 81 percent of the cases, the sign on the coef?cient was positive and statistically signi?cant. Indeed, for 6. See “Part I. The Country Chronologies and Chartbook, Background Material to A Modern History of Exchange Rate Arrangements: A Reinterpretation” at http://www.puaf.umd.edu/faculty/papers/reinhart/reinhart.htm. 7. Two de?nitions of currency crashes are used. A severe currency crash refers to a 25 percent or higher monthly depreciation which is at least 10 percent higher than the previous month’s depreciation. The “milder” version represents a 12.5 percent monthly depreciation which is at least 10 percent above the preceding month’s depreciation; see details in the Appendix. EXCHANGE RATE ARRANGEMENTS 7FIGURE II Of?cial Exchange Rates Typically Validate the Changes in the Market Rates Sources: Pick and Se´dillot [1971]; International Currency Analysis, World Currency Yearbook, various issues. 8 QUARTERLY JOURNAL OF ECONOMICSWestern Hemisphere as a region, the coef?cient on the parallel premium was signi?cant for all the countries in our sample. These ?ndings are in line with those of Bahmani-Oskooee, Miteza, and Nasir [2002], who use panel annual data for 1973– 1990 for 49 countries and employ a completely different approach. Their panel cointegration tests indicate that the of?cial rate will systematically adjust to the market rate in the long run. Second, we calculated pairwise correlations between ination (measured as the twelve-month change in the consumer price index) and the twelve-month percent change in the of?cial and market exchange rates, six months earlier. If the market rate is a better pulse of monetary policy, it should be (a priori) more closely correlated with ination. As shown in Table II, we ?nd that for the majority of cases (about three-quarters of the countries) the changes in market-determined exchange rates have higher correlations with ination than do changes in the of?cial rate. 8 An interesting exception to this pattern of higher correla- 8. Note that, due to data limitations, we use of?cial prices rather than black market or “street” prices to measure ination here. Otherwise, the dominance of the market-determined rates in this exercise would presumably be even more pronounced. TABLE I IS THE PARALLEL MARKET RATE A GOOD PREDICTOR OF CRASHES IN THE OFFICIAL EXCHANGE RATE? SUMMARY OF THE PROBIT COUNTRY-BY-COUNTRY ESTIMATION Regression, DO t 5 a 1 b DPt2 i 1 ut “Mild” crash Percent of countries for which: b . 0 97.1 b . 0 and signi?cant a 81.4 b , 0 2.9 b , 0 and signi?cant a 1.4 Sources: Pick’s Currency Yearbook, World Currency Report, Pick’s Black Market Yearbook, and the authors’ calculations. DOt is a dummy variable that takes on the value of 1 when there is a realignment in the of?cial exchange rate along the lines described below and 0 otherwise, a and b are the intercept and slope coef?cients, respectively (our null hypothesis is b . 0), DPt2i is the twelve-monthchange in the parallel exchange rate, lagged one to six months (the lags were allowed to vary country by country, as there was no prior reason to restrict dynamics to be the same for all countries) and ut is a random disturbance. Two de?nitions of currency crashes are used in the spirit of Frankel and Rose [1996]. A “severe” currency crash refers to a 25 percent or higher monthly depreciation, which is at least 10 percent higher than the previousmonth’s depreciation.The “mild” version represents a 12.5 percent monthly depreciation, which is at least 10 percent above the preceding month’s depreciation. Since both de?nitions of crash yield similar results, we report here only those for the more inclusive de?nition. The regression sample varies by country and is determined by data availability. a. At the 10 percent con?dence level or higher. EXCHANGE RATE ARRANGEMENTS 9tions between the market-determined exchange rate changes and ination is for the industrial countries in the “Convertible Bretton Woods” period (1959 –1973), an issue that merits further study. II.C. How Important Are Parallel Markets? There are cases where the parallel (or secondary) exchange rate applies only to a few limited transactions. An example is the “switch pound” in the United Kingdom during September 1950 through April 1967. 9 However, it is not unusual for dual or parallel markets (legal or otherwise) to account for the lion’s share of transactions with the of?cial rate being little more than symbolic. As Kiguel, Lizondo, and O’Connell [1997] note, the of?cial rate typically diminishes in importance when the gap between the of?cial and market-determined rate widens. To provide a sense of the comparative relevance of the dual or parallel market, we proceed along two complementary dimensions. First, we include a qualitative description in the countryspeci?c chronologies (see background material) of what transactions take place in the of?cial market versus the secondary market. Second, we develop a quantitative measure of the potential size of the leakages into dual or parallel exchange markets. 10 9. For example, while the United Kingdom of?cially had dual rates through April 1967, the secondary rate was so trivial (both in terms of the premium and the volume of transactions it applied to) that it is classi?ed as a peg in our classi?cation scheme (see background material). In the next section we describe how our classi?cation algorithm deals with these cases. 10. For instance, according to Claessens [1997], export underinvoicing hit a historic high in Mexico during 1982—the crisis year in which the dual market was TABLE II INFLATION, OFFICIAL AND MARKET-DETERMINED EXCHANGE RATES: COUNTRY-BY-COUNTRY PAIRWISE CORRELATIONS Percent of countries for which the correlations of: The market-determined exchange rate and ination are higher than the correlations of the of?cial rate and ination 73.7 The market-determined exchange rate and ination are lower than the correlations of the of?cial rate and ination 26.3 Sources: International Monetary Fund, International Financial Statistics, Pick’s Currency Yearbook, World Currency Report, Pick’s Black Market Yearbook, and the authors’ calculations. The correlations reported are those of the twelve-month percent change in the consumer price index with the twelve-month percent change in the relevant bilateral exchange rate lagged six months. 10 QUARTERLY JOURNAL OF ECONOMICSFollowing Ghei, Kiguel, and O’Connell [1997], we classify episodes where there are dual/parallel markets into three tiers according to the level (in percent) of the parallel market premium: low (below 10 percent), moderate (10 percent or above but below 50), and high (50 percent and above). For the episodes of dual/ parallel markets, we provide information about which category each episode falls in (by calculating the average premium for the duration of the episode). In addition to the information contained in the premium, we constructed an extensive database on export misinvoicing, or the difference between what a country reports as its exports and what other countries report as imports from that country, adjusted for shipping costs. Historically, there are tight links between capital ight, export underinvoicing, and the parallel market premium. 11 As with the parallel market premium, we divide the export misinvoicing estimates into three categories (as a percent of the value of total exports): low (less than 10 percent of exports), moderate (10 to 15 percent of exports), and high (above 15 percent). For Europe, Japan, and the United States, misinvoicing calculations start in 1948, while for the remaining countries these start in 1970. In the extensive background material to this paper, we show, for each episode, which of the three categories is applicable. Finally, we construct a score (1 for Low, 2 for Moderate, and 3 for High) for both of these proxies for leakages. The combined score on the estimated size of the leakages (these range from 2 to 6) is also reported. 12 Table III, which shows the evolution of export misinvoicing (as a percent of the value of total exports) and the parallel market premium (in percent) across regions and through time, provides a general avor of the size of potential leakages from the of?cial market. According to our estimates of misinvoicing (top panel), the regional patterns show the largest leakages for the Caribbean and non-CFA Sub-Saharan Africa 1970 –2001, with averages in the 30 to 50 percent range. The lowest estimates of misinvoicing (8 to 11 percent) are for Western Europe, North America, and the introduced. Similar statements can be made about other crisis episodes that involved the introduction of exchange controls and the segmentation of markets. 11. See Kiguel, Lizondo, and O’Connell [1997] and the references contained therein. 12. See “Part II. Parallel Markets and Dual and Multiple Exchange Rate Practices: Background Material to A Modern History of Exchange Rate Arrangements: A Reinterpretation” at http://www.puaf.umd.edu/faculty/papers/reinhart/reinhart.htm. EXCHANGE RATE ARRANGEMENTS 11TABLE III LEAKAGES: EXPORT MISINVOICING AND THE PARALLEL MARKET PREMIUM ABSOLUTE VALUE OF EXPORT MISINVOICING (AS A PERCENT OF THE VALUE OF EXPORTS) Descriptive statistics Mean absolute value (by decade) Min. Max. St. dev 48–49 50–59 60–69 70–79 80–89 90–01 70–01 World 7.0 39.8 8.4 12.8 10.9 9.9 24.7 22.1 26.0 24.4 North Africa 2.5 59.9 10.3 ... ... ... 7.2 8.3 16.1 10.9 CFA 12.6 48.3 8.4 ... ... ... 28.5 21.7 21.5 23.8 Rest of Africa 16.3 201.9 33.5 ... ... ... 23.4 23.4 53.6 34.1 Middle East and Turkey 9.1 45.4 9.6 ... ... ... 30.7 16.7 17.4 21.5 Developing Asia and Paci?c 9.5 79.1 16.9 ... ... ... 31.4 14.9 24.1 23.5 Industrialized Asia 3.7 18.2 3.3 11.2 14.2 13.9 14.6 12.0 10.3 12.2 Caribbean 9.7 136.0 33.2 ... ... ... 30.8 48.9 60.0 47.0 Central and South America 12.0 49.6 8.2 ... ... ... 26.1 36.0 30.4 30.8 Central and Eastern Europe 2.5 50.0 18.3 ... ... ... 46.6 15.4 7.4 22.1 Western Europe 2.4 16.9 3.0 14.1 10.4 10.0 11.6 7.6 7.7 8.9 North America 0.6 22.6 5.9 4.6 9.4 3.8 16.0 11.4 4.8 10.4 Monthly average parallel market premium (excluding freely falling episodes, in percent) Descriptive statistics Average (by decade) Min. Max. St. dev 46–49 50–59 60–69 70–79 80–89 90–98 46–98 World 11.6 205.9 35.4 137.8 56.7 38.1 31.3 57.8 52.6 54.1 North Africa 21.2 164.8 41.4 ... 9.9 35.7 30.7 108.6 62.0 53.6 CFA 26.4 12.7 2.7 ... ... ... 0.0 1.2 1.8 0.9 Rest of Africa 1.7 322.5 73.9 31.9 6.9 33.7 113.7 112.7 107.7 71.0 Middle East and Turkey 5.1 493.1 99.6 54.6 81.0 26.0 21.4 146.5 193.2 88.6 Developing Asia and Paci?c 23.7 660.1 95.0 143.5 60.9 168.9 44.7 43.1 12.1 72.9 Industrialized Asia 26.9 815.9 107.6 324.4 43.0 12.0 3.6 1.3 1.5 36.1 Caribbean 223.8 300.0 42.8 ... ... 29.6 30.2 56.8 53.6 42.3 Central and South America 3.0 716.1 78.5 49.1 133.0 16.4 18.6 74.8 8.4 51.0 Western Europe 25.6 347.5 48.6 165.5 17.0 1.2 2.0 1.7 1.2 16.9 North America 24.3 49.7 3.3 7.2 0.5 0.0 1.1 1.4 1.6 1.3 Sources: International Monetary Fund, Direction of Trade Statistics, International Financial Statistics, Pick’s Currency Yearbook, World Currency Report, Pick’s Black Market Yearbook, and authors’ calculations. To calculate export misinvoicing, let XWi 5 imports from country i, as reported by the rest of the world (CIF basis), Xi 5 exports to the world as reported by country i, Z 5 imports CIF basis/imports COB basis, then export misinvoicing 5 (XWi /Z) 2 Xi . The averages reported are absolute values as a percent of the value of total exports. The parallel premium is de?ned as 100 3 [(P 2 O)/O)], where P and O are the parallel and of?cial rates, respectively. The averages for the parallel premium are calculated for all the countries in our sample in each region, as such, it includes countries where rates are uni?ed and the premium is zero or nil. 12 QUARTERLY JOURNAL OF ECONOMICSCFA Franc Zone. It is also noteworthy that, although low by the standards of other regions, the export misinvoicing average in 1970 –2001 for Western Europe is half of what it was in 1948 – 1949. Yet these regional averages may understate the importance of misinvoicing in some countries. For example, the maximum value for 1948 –2001 for Western Europe (16.9 percent) does not reect the fact that for Spain misinvoicing as a percent of the value of exports amounted to 36 percent in 1950, a comparable value to what we see in some of the developing regions. As to the regional average parallel market premium shown in the bottom panel of Table III, all regions fall squarely in the Moderate-to-High range (with the exception of North America, Western Europe, and CFA Africa). In the case of developing Asia, the averages are signi?cantly raised by Myanmar and Laos. It is worth noting the averages for Europe and industrialized Asia in the 1940s are comparable and even higher than those recorded for many developing countries, highlighting the importance of acknowledging and accounting for dual markets during this period. To sum, in this section we have presented evidence that leads us to conclude that parallel markets were both important as indicators of monetary policy and as representative of the prices underlying an important share of economic transactions. It is therefore quite reasonable to draw heavily on the dual or parallel market data in classifying exchange rate regimes, the task to which we now turn. III. THE “NATURAL” CLASSIFICATION CODE: A GUIDE We would describe our classi?cation scheme as a “Natural” system that relies on a broad variety of descriptive statistics and chronologies to group episodes into a much ?ner grid of regimes, rather than the three or four buckets of other recent classi?cation strategies. 13 The two most important new pieces of information we bring to bear are our extensive data on market-determined dual or parallel exchange rates and detailed country chronologies. The data, its sources, and country coverage are described along with the chronologies that map the history of exchange rate arrangements for each country in the detailed background mate- 13. In biology, a natural taxonomic scheme relies on the characteristics of a species to group them. EXCHANGE RATE ARRANGEMENTS 13rial to this paper. To verify and classify regimes, we also rely on a variety of descriptive statistics based on exchange rate and ination data from 1946 onwards; the Appendix describes these. III.A. The Algorithm Figure III is a schematic summarizing our Natural Classi?- cation algorithm. First, we use the chronologies to sort out for separate treatment countries with either of?cial dual or multiple rates or active parallel (black) markets. 14 Second, if there is no dual or parallel market, we check to see if there is an of?cial preannounced arrangement, such as a peg or band. If there is, we examine summary statistics to verify the announced regime, going forward from the date of the announcement. If the regime is veri?ed (i.e., exchange rate behavior accords with the preannounced policy), it is then classi?ed accordingly as a peg, crawling peg, etc. If the announcement fails veri?cation (by far the most common outcome), we then seek a de facto statistical classi?cation using the algorithm described below, and discussed in greater detail in the Appendix. Third, if there is no preannounced path for the exchange rate, or if the announced regime cannot be veri?ed by the data and the twelve-month rate of ination is below 40 percent, we classify the regime by evaluating exchange rate behavior. As regards which exchange rate is used, we consider a variety of potential anchor currencies including the US dollar, deutsche mark, euro, French franc, UK pound, yen, Australian dollar, Italian lira, SDR, South African rand, and the Indian rupee. A reading of the country chronologies makes plain that the relevant anchor currency varies not only across countries but sometimes within a country over time. (For example, many former British colonies switched from pegging to the UK pound to pegging to the US dollar.) Our volatility measure is based on a ?ve-year moving window (see the Appendix for details), so that the monthly exchange rate behavior may be viewed as part of a larger, continuous, regime. 15 14. See background material posted at http://www.puaf.umd.edu/faculty/ papers/reinhart/reinhart.htm. 15. If the classi?cation is based on exchange rate behavior in a particular year, it is more likely that one-time events (such as a one-time devaluation and repeg) or an economic or political shock leads to labeling the year as a change in regime, when in effect there is no change. For example, Levy-Yeyati and Sturzenegger [2002], who classify regimes one year at a time (with no memory), classi?ed all CFA zone countries as having an intermediate regime in 1994, when 14 QUARTERLY JOURNAL OF ECONOMICSthese countries had a one-time devaluation in January of that year. Our algorithm classi?es them as having pegs throughout. The ?ve-year window also makes it less likely that we classify as a peg an exchange rate that did not move simply because it was a tranquil year with no economic or political shocks. It is far less probable that there are no shocks over a ?ve-year span. FIGURE III A Natural Exchange Rate Classi?cation Algorithm EXCHANGE RATE ARRANGEMENTS 15We also examined the graphical evidence as a check on the classi?cation. In practice, the main reason for doing so is to separate pegs from crawling pegs or bands and to sort the latter into crawling and noncrawling bands. Fourth, as we have already stressed, a straightforward but fundamental departure from all previous classi?cation schemes is that we create a new separate category for countries whose twelve-month rate of ination is above 40 percent. These cases are labeled “freely falling.” 16 If the country is in a hyperination (according to the classic Cagan [1956] de?nition of 50 percent or more monthly ination), we categorize the exchange rate regime as a “hyperoat,” a subspecies of freely falling. In Figure IV, bilateral exchange rates versus the US dollar are plotted for two countries that have been classi?ed by the IMF (and all previous classi?cation efforts) as oating over much of the postwar period—Canada and Argentina. 17 To us, lumping the Canadian oat with that of Argentina during its hyperination seems, at a minimum, misleading. As Figure IV illustrates, oating regimes look rather different from freely falling regimes—witness the orders of magnitude difference in the scales between Canada (top of page) and Argentina (bottom). This difference is highlighted in the middle panel, which plots the Canadian dollar-US dollar exchange rate against Argentina’s scale; from this perspective, it looks like a ?xed rate! The exchange rate histories of other countries that experienced chronic high ination bouts—even if these did not reach the hyperination stage—look more similar to Argentina in Figure IV than to Canada. 18 In our view, regimes associated with an utter lack of monetary control and the attendant very high ination should not be automatically lumped under the same exchange rate arrangement as low ination oating regimes. On these grounds, freely falling needs to be treated as a separate category, much in the same way that Highly Indebted Poorest Countries (HIPC) are treated as a separate “type” of debtor. 16. In the exceptional cases (usually the beginning of an ination stabilization plan) where, despite ination over 40 percent, the market rate nevertheless follows a con?rmed, preannounced band or crawl, the preannounced regime takes precedence. 17. For Argentina, this of course refers to the period before the Convertibility Plan is introduced in April 1991 and for Canada the post-1962 period. 18. Two-panel ?gures, such as that shown for Chile (Figure V), for each country in the sample are found in the background material alongside the country-speci?c chronologies. 16 QUARTERLY JOURNAL OF ECONOMICSFIGURE IV The Essential Distinction between Freely Floating and Falling Sources: Pick and Se´dillot [1971]; International Currency Analysis, World Currency Yearbook, various issues. EXCHANGE RATE ARRANGEMENTS 17In step 5 we take up those residual regimes that were not classi?ed in steps 1 through 4. These regimes become candidates for “managed” or “freely” oating. 19 To distinguish between the two, we perform some simple tests (see the Appendix) that look at the likelihood the exchange rate will move within a narrow range, as well as the mean absolute value of exchange rate changes. When there are dual or parallel markets and the parallel market premium is consistently 10 percent or higher, we apply steps 1 through 5 to our data on parallel exchange rates and reclassify accordingly, though in our ?ner grid. 20 III.B. Using the Chronologies The 153 individual country chronologies are also a central point of departure from all previous efforts to classify regimes. In the ?rst instance the data are constructed by culling information from annual issues of various secondary sources, including Pick’s Currency Yearbook, World Currency Yearbook, Pick’s Black Market Yearbook, International Financial Statistics, the IMF’s Annual Report on Exchange Rate Arrangements and Exchange Restrictions, and the United Nations Yearbook. Constructing our data set required us to sort and interpret information for every year from every publication above. Importantly, we draw on national sources to investigate apparent data errors or inconsistencies. More generally, we rely on the broader economics literature to include pertinent information, such as the distribution of transactions among of?cial and parallel markets. 21 The chronologies allow us to date dual or multiple exchange rate episodes, as well as to differentiate between preannounced pegs, crawling pegs, and bands from their de facto counterparts. We think it is important to distinguish between, say, de facto pegs or bands from announced pegs or bands, because their properties are potentially different. 22 At the very least, we want to provide future researchers with the data needed to ask a variety of questions about the role of exchange rate arrangements. The 19. Our classi?cation of “freely oating” is the analogue of “independently oating” in the of?cial classi?cation. 20. When the parallel market premium is consistently (i.e., all observations within the ?ve-year window) in single digits, we ?nd that in nearly all these cases the of?cial and parallel rates yield the same classi?cation. 21. See Marion [1994], for instance. 22. Policy-makers may not be indifferent between the two. In theory, at least, announcements of pegs, bands, and so on can act as a coordinating device which, by virtue of being more transparent, could invite speculative attacks. 18 QUARTERLY JOURNAL OF ECONOMICSchronologies also ag the dates for important turning points, such as when the exchange rate ?rst oated, or when the anchor currency was changed. Table IV gives an example of one of our 153 chronologies (see background material) for the case of Chile. The ?rst column gives critical dates. Note that we extend our chronologies as far back as possible (even though we can only classify from 1946 onwards); in the case of Chile we go back to 1932. The second column lists how the arrangement is classi?ed. Primary classi?cation refers to the classi?cation according to our Natural algorithm, which may or may not correspond to the of?cial IMF classi?cation (shown in parentheses in the second column of Table IV). Secondary and tertiary classi?cations are meant only to provide supplemental information, as appropriate. So, for example, from November 1952 until April 1956, Chile’s ination was above 40 percent, and hence, its primary classi?cation is freely falling—that is, the only classi?cation that matters for the purposes of the Natural algorithm. For those interested in additional detail, however, we also note in that column that the market-determined exchange rate was a managed oat along the lines described in detail in the Appendix (secondary) and that, furthermore, Chile had multiple exchange rates (tertiary). This additional information may be useful, for example, for researchers who are not interested in treating the high ination cases separately (as we have done here). In this case, they would have suf?cient information to place Chile in the 1952–1956 period in the managed oat category. Alternatively, for those researchers who wish to treat dual or multiple exchange rate practices as a separate category altogether (say, because these arrangements usually involve capital controls), the second column (under secondary or tertiary classi?cation) provides the relevant information to do that sorting accordingly. As one can see, although Chile uni?ed rates on September 1999, it previously had some form of dual or multiple rates throughout most of its history. In these circumstances, we reiterate that our classi?cation algorithm relies on the market-determined, rather than the of?cial exchange rate. 23 Over some 23. The other Chronologies do not contain this information, but the annual of?cial IMF classi?cation for the countries in the sample is posted at http:// www.puaf.umd.edu/faculty/papers/reinhart/reinhart.htm. EXCHANGE RATE ARRANGEMENTS 19TABLE IV A SAMPLE CHRONOLOGY NI THE NATURAL CLASSIFICATION SCHEME: CHILE, 1932–2001 Date Classi?cation pr mi ary/secondary/tertiary (of?cial MI F classi?cation in parentheses) Comments September 16, 1925–April 20, 1932 Peg Go dl standard. Foreign exchange controls are ni troduced on July 30, 1931. April 20, 1932–1937 Dual market Pound Sterl ni g is reference cu ency. Suspension of go dl standard. 1937–February 1946 Managed oat ni g M/ ult pi le rates US dol al r becomes the reference cu ency. March 1946–May 1947 Freely af ll ni g M/ anaged oat ni g M/ ult pi le rates June 1947–October 1952 Managed oat ni g M/ ult pi le rates November 1952–April 16, 1956 Freely af ll ni g M/ anaged oat ni g M/ ult pi le rates April 16, 1956–August 1957 Freely af ll ni g M/ anaged oat ni g D/ ual market Rate structure is s mi pl ?i ed, and a dual market is created. September 1957–June 1958 Managed oat ni g D/ ual market July 1958–January 1, 1960 Freely af ll ni g M/ anaged oat ni g D/ ual market January 1, 1960–January 15, 1962 Peg to US dol al r The escudo rep al ces the peso. January 15, 1962–November 1964 Freely af ll ni g M/ anaged oat ni g M/ ult pi le rates Freely af ll ni g s ni ce April 1962. December 1964–June 1971 Managed oat ni g M/ ult pi le rates P( eg) July 1971–June 29, 1976 Freely af ll ni g M/ ult pi le exchange rates P( eg through 1973 m- anaged oat ni g afte wr ards) On September 29, 1975, the peso rep al ced the escudo. October 1973 c al ss ?i es as a hyperoat. June 29, 1976–January 1978 Freely af ll ni g C/ rawl ni g peg to US dol al r M( anaged oat ni g) February 1978–June 1978 Preannounced crawl ni g peg to US dol al r F/ reely fall ni g M( anaged oat ni g) The Tablita P al n. July 1978–June 30, 1979 Preannounced crawl ni g peg to US dol al r P( eg) The Tablita P al n. June 30, 1979–June 15, 1982 Peg to US dol al r P( eg) The second phase of the Tablita P al n. June 15, 1982–December 1982 Freely af ll ni g M/ anaged oat ni g D/ ual market January 1983–December 8, 1984 Managed oat ni g D/ ual market M( anaged oat ni g) Parallel market prem ui m reaches 102 percent ni early 1983. On March 1983 the ni tention to follow a P rule was announced. 20 QUARTERLY JOURNAL OF ECONOMICSTABLE IV (CONT NI UED) Date Classi?cation pr mi ary/secondary/tertiary (of?cial MI F classi?cation in parentheses) Comments December 8, 1984–January 1988 Managed oat ni g D/ ual market M( anaged oat ni g) rule. The o ?f c ai l rate is kept with ni a 62% crawl ni g band to US dol al r. February 1988–January 1, 1989 De facto crawl ni g band around US dol al r D/ ual market M( anaged oat ni g) rule. 65% band. O ?f c ai l preannounced 63% crawl ni g band to US dol al r. While the of?c ai l rate rema ni s with ni the preannounced band, parallel market prem ui m rema ni ni double digits. January 1, 1989–January 22, 1992 Preannounced crawl ni g band around US dol al r D/ ual market M( anaged oat ni g) rule. Band w di th is 65%. January 22, 1992–January 20, 1997 De facto crawl ni g band around US dol al r D/ ual market M( anaged oat ni g) rule. Band is 65%. There is an o ?f c ai l preannounced 610% crawl ni g band to US dol al r. Parallel prem ui m falls below 15 percent and ni to s ni gle digits. January 20, 1997–June 25, 1998 De facto crawl ni g band to US dol al r D/ ual market M( anaged oat ni g) Of?c ai l prea ounced crawl ni g 612 5. % band to US dol al r; de facto band is 65%. June 25, 1998–September 16, 1998 Preannounced crawl ni g band to US dol al r D/ ual market M( anaged oat ni g) 62 7. 5% band. September 16, 1998–December 22, 1998 Preannounced crawl ni g band to US dol al r D/ ual market M( anaged oat ni g) 63 5. % band. December 22, 1998–September 2, 1999 Preannounced crawl ni g band to US dol al r D/ ual market M( anaged oat ni g) 68% band. September 2, 1999–December 2001 Managed oat ni g (Independently oat ni g) Rates are un ?i ed. Reference currency is the US dollar. Data availability: Of?cial rate, 1900:1–2001:12. Parallel rate, 1946:1–1998:12. EXCHANGE RATE ARRANGEMENTS 21periods the discrepancy between the of?cial and parallel rate, however, proved to be small. For example, from January 1992 onwards the parallel market premium remained in single digits, and our algorithm shows that it makes little difference whether the of?cial or parallel rate is used. In these instances, we leave the notation in the second column that there are dual rates (for information purposes), but also note in the third column that the premium is in single digits. As noted, Chile has also experienced several periods where the twelve-month monthly ination exceeded 40 percent. Our algorithm automatically categorizes these as freely falling exchange rate regimes—unless there is a preannounced peg, crawling peg, or narrow band that is veri?ed, as was the case when the Tablita program was introduced on February 1978. The third column in our chronology gives further sundry information on the regime—e.g., the width of the announced and de facto bands, etc. For Chile, which followed a crawling band policy over many subperiods, it is particularly interesting to note the changes over time in the width of the bands. The third column also includes information about developments in the parallel market premium and currency reform. As an example of the former, we note that since 1992 the parallel premium slipped into single digits; an example of the latter is given for Chile when the peso replaced the escudo in 1975. The top panel of Figure V plots the path of the of?cial and market-determined exchange rate for Chile from 1946. It is evident that through much of the period shown the arrangement was one of a crawling peg or a crawling band, with the rate of crawl varying through time and notably slowing as ination began to stabilize following the Tablita plan of the early 1980s. The bottom panel plots the parallel market premium (in percent). This pattern is representative of many other countries in our sample; the premium skyrockets in the periods of economic and political instability, declines into single digits as credible policies are put in place and capital controls are eased. As we will discuss in the next section, the Chilean case is also illustrative, in that crawling pegs or bands are quite common. Figure VI, which shows the path of the exchange rate for the Philippines, India, and Greece, provides other examples of the plethora of crawling pegs or bands in our sample. 22 QUARTERLY JOURNAL OF ECONOMICSFIGURE V Chile: Of?cial and Market-Determined Exchange Rates and the Parallel Market Premium January 1946–December 1998 Sources: InternationalMonetary Fund, Annual Report on Exchange Arrangements and Exchange Restrictions and International Financial Statistics; Pick and Se´dillot [1971]; International Currency Analysis, World Currency Yearbook, various issues. EXCHANGE RATE ARRANGEMENTS 23FIGURE VI The Prevalence of Crawling Pegs and Bands Sources: Pick and Se´dillot [1971]; International Currency Analysis, World Currency Yearbook, various issues. 24 QUARTERLY JOURNAL OF ECONOMICSIII.C. Alternative Taxonomies: Comparing the Basic Categories Altogether, our taxonomy of exchange rate arrangements includes the fourteen classi?cations sketched in Table V (or ?fteen if hyperoats are treated as a separate category). Of course, fourteen (or ?fteen) buckets are not exhaustive, for example, if one wishes to distinguish between forward- and backward-looking crawls or bands, along the lines of Cottarelli and Giannini [1998]. Given that we are covering the entire post-World War II period, we did not have enough information to make that kind of ?ner distinction. Conversely, because we sometimes want to compare our classi?cation regime with the coarser of?cial one, we also show how to collapse our fourteen types of arrangements into ?ve broader categories; see Table V, where the least exible arrangements are assigned the lowest values in our scale. TABLE V THE FINE AND COARSE GRIDS OF THE NATURAL CLASSIFICATION SCHEME Natural classi?cation bucket Number assigned to category in: Fine grid Coarse grid No separate legal tender 1 1 Preannounced peg or currency board arrangement 2 1 Preannounced horizontal band that is narrower than or equal to 62% 3 1 De facto peg 4 1 Preannounced crawling peg 5 2 Preannounced crawling band that is narrower than or equal to 62% 6 2 De facto crawling peg 7 2 De facto crawling band that is narrower than or equal to 62% 8 2 Preannounced crawling band that is wider than 62% 9 2 De facto crawling band that is narrower than or equal to 65% 10 3 Noncrawling band that is narrower than or equal to 62% a 11 3 Managed oating 12 3 Freely oating 13 4 Freely falling (includes hyperoat) 14 5 Source: The authors. a. By contrast to the common crawling bands, a noncrawling band refers to the relatively few cases that allow for both a sustained appreciation and depreciation of the exchange rate over time. While the degree of exchange rate variability in these cases is modest at higher frequencies (i.e., monthly), lower frequency symmetric adjustment is allowed for. The Appendix provides a detailed discussion of our classi?cation algorithm. EXCHANGE RATE ARRANGEMENTS 25In the ?ner grid, we distinguish between preannounced policies and the less transparent de facto regimes. Since the former involve an explicit announcement while the latter leave it to ?nancial market analysts to determine the implicit exchange rate policy, in the ?ner classi?cation we treat preannouncement as less exible than de facto. We accordingly assign it a lower number in our scale. Those not interested in testing whether announcements serve as a coordinating device (say, to make a speculative attack more likely) and only interested in sorting out the degree of observed exchange rate exibility will prefer the coarser grid. However, even in the coarse grid, it is imperative to treat freely falling as a separate category. IV. THE “NATURAL” TAXONOMY: CRITIQUES AND COMPARISONS As the previous section described, our classi?cation strategy relies importantly on the observed behavior of the market-determined exchange rate. In this section we ?rst address some potential critiques of our approach, including whether a country’s international reserve behavior should affect its classi?cation, and whether we may be mislabeling some regimes as pegs or crawls simply due to the absence of shocks. We then proceed to compare our results with the “of?cial history,” and provide examples of how our reclassi?cation may reshape some of the existing evidence on the links between exchange rate arrangements and various facets of economic activity. IV.A. The Trilogy: Exchange Rates, Monetary Policy, and Capital Controls To capture the nuances of any exchange rate arrangement, one might also want information on the presence and effectiveness of capital controls, the modalities of (sterilized or unsterilized) foreign exchange intervention, and the extent to which interest rates (or other less conventional types of intervention) are used as a means to stabilize the exchange rate. Since, for the purposes of universality, our classi?cation rests squarely on the univariate time series behavior of the nominal exchange rates (combined with historical chronologies), in this subsection we address some of these limitations to our approach. Some studies have reclassi?ed exchange rate arrangements by also factoring in the behavior of foreign exchange reserves as 26 QUARTERLY JOURNAL OF ECONOMICSreported by the IMF’s International Financial Statistics. 24 However, as Calvo and Reinhart [2002] note, using reserves has serious limitations. In Brazil and in over two dozen other countries, foreign exchange market intervention is frequently done through purchases and sales of domestic dollar-linked debt. 25 This debt is not reected in the widely used IFS reserve data, neither were the massive interventions of the Thai authorities in the forward market during 1997 and in South Africa thereafter. Furthermore, as ?nancial liberalization has spread throughout the globe, there has been a widespread switch from direct intervention in the foreign exchange market to the use of interest rate policy in the 1990s as a means to stabilize the exchange rate. 26 Picking up on this kind of policy intervention requires having the policy interest rate—the equivalent of the federal funds rate for the United States—for each country. Such data are very dif?cult to come by, and none of the other efforts at reclassi?cation have dealt with issue. Other issues arise in the context of the links between monetary, capital controls, and exchange rate policy. In particular, while ?xing the exchange rate (or having narrow bands, or crawling pegs, or bands) largely de?nes monetary policy, our two most exible arrangement categories (managed or freely oating) do not. Floating could be consistent with monetary targets, interest rate targets, or ination targeting, the latter being a relatively recent phenomenon. 27 Since our study dates back to 1946, it spans a sea change in capital controls and monetary policy regimes, and it is beyond the scope of this paper to subdivide the monetary policy framework for the most exible arrangements in 24. For instance, the algorithm used by Levy-Yeyati and Sturzenegger [2002] also uses (besides the exchange rate) reserves and base money. This gives rise to many cases of what they refer to as “one classi?cation variable not available.” This means that their algorithm cannot provide a classi?cation for the United Kingdom (where it is hard to imagine such data problems) until 1987 and—in the most extreme of cases—some developing countries cannot be classi?ed for any year over their 1974–2000 sample. 25. See Reinhart, Rogoff, and Savastano [2003] for a recent compilation of data on domestic dollar-linked debt. 26. There are plenty of recent examples where interest rates were jacked up aggressively to fend off a sharp depreciation in the currency. Perhaps one of the more obvious examples is in the wake of the Russian default in August 1998, when many emerging market currencies came under pressure and countries like Mexico responded by doubling interest rates (raising them to 40 percent) within a span of a couple of weeks. 27. Indeed, several of the ination targeters in our sample (United Kingdom, Canada, Sweden, etc.) are classi?ed as managed oaters. (However, it must also be acknowledged that there are many different variants of ination targeting, especially in emerging markets.) EXCHANGE RATE ARRANGEMENTS 27our grid. Apart from exchange rate policy, however, our study sheds considerable light on the third leg of the trinity—capital controls. While measuring capital mobility has not been the goal of this paper, our data consistently show that the parallel market premium dwindles into insigni?cance with capital market integration, providing a promising continuous measure of capital mobility. IV.B. Exchange Rates and Real Shocks Ideally, one would like to distinguish between exchange rate stability arising from deliberate policy actions (whether its direct foreign exchange market intervention or interest rate policy, as discussed) and stability owing to the absence of economic or political shocks. In this subsection we provide evidence that, if the exchange rate is stable and it is accordingly treated in our de jure approach to classi?cation, it is typically not due to an absence of shocks. Terms of trade shocks are a natural source of potential shocks, particularly for many developing countries. Similarly, the presence (or absence) of shocks is likely to be reected in the volatility of real GDP. To investigate the incidence and size of terms of trade shocks, we constructed monthly terms of trade series for 172 countries over the period 1960 –2001. 28 The terms of trade series is a geometric weighted average of commodity prices (?xed weights based on the exports of 52 commodities). Table VI presents a summary by region of the individual country ?ndings. The ?rst column shows the share of commodities in total exports, while s Dtot denotes the variance of the monthly change in the terms of trade of the particular region relative to Australia. Australia is our benchmark, as it is both a country that is a primary commodity exporter and has a oating exchange rate that, by some estimates, approximates an optimal response to terms of trade shocks (see Chen and Rogoff [2003]). The next three columns show the variance of the monthly change in the terms of trade of the region relative to Australia (s Dtot), exchange rate of the individual region relative to Australia (s De) and the variance of the annual change in real GDP of the region relative to Australia (s Dy). The last two columns show the 28. Table VI is based on the more extensive results in Reinhart, Rogoff, and Spilimbergo [2003]. 28 QUARTERLY JOURNAL OF ECONOMICSvariance of the exchange rate relative to the variance of the terms of trade (s De)/(s Dtot) and output (s De)/(s Dy), respectively. A priori, adverse terms of trade shocks should be associated with depreciations and the converse for positive terms of trade shocks; greater volatility in the terms of trade should go handin-hand with greater volatility in the exchange rate. (In Chen and Rogoff [2003] there is greater volatility even under optimal policy.) Table VI reveals several empirical regularities: (a) most countries (regions) have more variable terms of trade than Australia—in some cases, such as the Middle East and the Caribbean, as much as three or four times as variable; (b) real GDP is also commonly far more volatile than in Australia; (c) most countries’ exchange rates appear to be far more stable than Australia’s, as evidenced by relatively lower variances for most of the groups; (d) following from the previous observations, the last two columns show that for most of the country groupings that the variance of exchange rate changes is lower than that of changes in the terms of trade or real GDP. Taken together, the implication of these ?ndings is that if the exchange rate is not moving, it is TABLE VI TERMS OF TRADE, OUTPUT, AND EXCHANGE RATE VARIABILITY VARIANCE RATIOS (NORMALIZED TO AUSTRALIA AND EXCLUDES FREELY FALLING EPISODES) Region Share s Dtot s De s Dy s De s Dtot s De s Dy North Africa 0.51 3.29 0.93 2.54 0.64 0.23 Rest of Africa (excluding CFA) 0.56 2.92 2.87 2.50 1.29 1.38 Middle East 0.60 4.15 0.95 3.48 0.33 0.50 Development Asia/Paci?c 0.34 2.02 0.85 2.40 0.54 0.44 Industrialized Asia 0.18 0.82 0.97 1.15 1.23 0.86 Caribbean 0.50 4.15 0.67 2.40 0.20 0.35 Central America 0.62 3.02 0.49 2.11 0.21 0.28 South America 0.63 2.03 1.08 2.15 0.66 0.52 Central East Europe 0.24 0.60 1.03 1.51 1.66 0.78 Western Europe 0.18 1.75 0.84 1.25 0.76 0.56 North America 0.33 1.64 0.60 1.12 0.47 0.54 Source: Reinhart, Rogoff, and Spilimbergo [2003] and sources cited therein. The variable de?nitions are as follows: Share 5 share of primary commodities to total exports; the next three columns show the variance of the monthly change in the terms of trade of the region relative to Australia (s Dtot), the variance of the monthly change in the exchange rate of the individual region relative to Australia (s De), and the variance of the annual change in real GDP of the region relative to Australia (s Dy); the last two columns show the variance of the exchange rate relative to the variance of the terms of trade (s De)/(s Dtot) and output (s De)/(s Dy), respectively. EXCHANGE RATE ARRANGEMENTS 29not for lack of shocks. Of course, terms of trade are only one class of shocks that can cause movement in the exchange rate. Thus, considering other kinds of shocks—political and economic, domestic, and international—would only reinforce the results presented here. IV.C. Fact and Fiction: Natural and Arti?cial? We are now prepared to contrast the of?cial view of the history of exchange rate regimes with the view that emerges from employing our alternative methodology. To facilitate comparisons, we will focus mainly on the coarse grid version of the Natural system. Figure VII highlights some of the key differences between the Natural and IMF classi?cations. The dark portions of the bars denote the cases where there is overlap between the IMF and the Natural classi?cation. 29 The white bar shows the cases where the IMF labels the regime in one way (say, a peg in 1970 –1973) and the Natural labels it differently. Finally, the striped portions of the bars indicate the cases where the Natural classi?cation labels the regime in one way (say, freely falling, 1991–2001) and the IMF labels differently (say, freely oating). As shown in Figure VII, according to our Natural classi?cation system, about 40 percent of all regimes in 1950 were pegs (since many countries had dual/parallel rates that did not qualify as pegs). Figure VII also makes plain that some of the “pegs” in our classi?cation were not considered pegs under the of?cial classi?cation; in turn, our algorithm rejects almost half of the of?cial pegs as true pegs. Our reclassi?cation of the early postwar years impacts not only on developing countries, but on industrialized countries as well; nearly all the European countries had active parallel markets after World War II. A second reason why our scheme shows fewer pegs is that the IMF’s pre-1997 scheme allowed countries to declare their regimes as “pegged to an undisclosed basket of currencies.” This notably nontransparent practice was especially popular during the 1980s, and it was also under this that a great deal of managed oating, freely oating, and freely falling actually took place. For the period 1974 –1990 the of?cial classi?cation has roughly 60 percent of all regimes as pegs; our classi?cation has only half as many. Again, as we see in Figure VII, this comparison 29. Speci?cally, both classi?cations assigned the regime for a particular country in a given particular year to the same category. 30 QUARTERLY JOURNAL OF ECONOMICSunderstates the differences since some of our pegs are not of?cial pegs and vice versa. For the years 1974 –1990, and 1991–2001, one can see two major trends. First, “freely falling” continues to be a signi?cant category, accounting for 12 percent of all regimes from 1974 –1990, and 13 percent of all regimes from 1991–2001. For the transition economies in the 1990s, over 40 percent of the observations are in the freely falling category. Of course, what we are reporting in Figure VII is the incidence of each regime. Clearly, future research could use GDP weights and—given that FIGURE VII Comparison of Exchange Rate Arrangements According to the IMF Of?cial and Natural Classi?cations, 1950–2001 Sources: International Monetary Fund, Annual Report on Exchange Arrangements and Exchange Restrictions and International Financial Statistics; Pick and Se´dillot [1971]; International Currency Analysis, World Currency Yearbook, various issues. The dark bars show the overlap between the IMF and Natural classi?cation (i.e., for that particular year the IMF and Natural classi?cations coincide); the white bars show the cases where the IMF classi?cation labeled the regime in one way (say, a peg in 1974–1990) and the Natural classi?cation labeled it differently; the striped bars indicate the cases where the Natural classi?cation labeled the regime in one way (say, freely falling) and the IMF labeled it differently, (say, freely oating). EXCHANGE RATE ARRANGEMENTS 31low-income countries are disproportionately represented in the freely falling category—this would reveal a lower importance to this category. 30 Second, the Natural classi?cation scheme reveals a bunching to the middle in terms of exchange rate exibility, when compared with the of?cial monetary history of the world. Limited exibility—which under the Natural classi?cation is dominated by de facto crawling pegs—becomes notably more important. From being a very small class under the of?cial scheme, the Natural classi?cation algorithm elevates limited exibility to the second most important grouping over the past decade, just behind pegs. Another startling difference is the reduced importance of freely oating. According to the of?cial classi?cation, more than 30 percent of countries were independently oating during 1991– 2001. According to the Natural classi?cation, less than 10 percent were freely oating. This is partly a manifestation of what Calvo and Reinhart [2002] term “fear of oating,” but equally because we assign high ination oats (including ones that are of?cially “pegs”) to our new freely falling category. Indeed, more countries had freely falling exchange rates than had freely oating exchange rates! The contrast between the IMF and Natural classi?cation systems becomes even more striking when one sees just how small the overlap is between the two classi?cations country by country and year by year. As shown in Table VII, if the IMF designation of the regime is a peg (1970 –2001), there is a 44 percent probability that our algorithm will place it into a more exible arrangement. If the of?cial regime is a oat, there is a 31 percent chance we will categorize it as a peg or limited exibility. If the of?cial regime is a managed oat, there is a 53 percent chance our algorithm will categorize it as a peg or limited exibility. Whether the of?cial regime is a oat or peg, it is virtually a coin toss whether the Natural algorithm will yield the same result. The bottom of the table gives the pairwise correlation between the two classi?cations, with the of?cial classi?cation running from 1 (peg) to 4 (independently oating), and the Natural classi?cation running from 1 (peg) to 5 (freely falling). The simple correlation coef?cient is only 0.42. As one can con?rm from 30. GDP weights and population weights would, of course, present very different pictures. For example, the United States and Japan alone would increase the world’s share of oaters if it were GDP weights, while weight by population would increase the weight of ?xers by China alone. 32 QUARTERLY JOURNAL OF ECONOMICSthe chronologies, the greatest overlap occurs in the classi?cation of the G3 currencies and of the limited exibility European arrangements. Elsewhere, and especially in developing countries, the two classi?cations differ signi?cantly, as we shall see. IV.D. The Pegs That Float Figure VIII plots the parallel market premium since January 1946, in percent, for Africa, Asia, Europe, and Western Hemisphere. As is evident from the Figure VIII, for all the regions except Europe, it would be dif?cult to make the case that the breakdown of Bretton Woods was a singular event, let alone a sea change. 31 For the developing world, the levels of pre- and post-1973 volatilities in the market-determined exchange rate, as revealed by the parallel market premium, are remarkably similar. Note that for all regions, we exclude the freely falling episodes that would signi?cantly increase the volatility but also distort the scale. To give a avor of the cross-country variation within region and across time, the dashed line plots the regional average plus one standard deviation (calculated across countries and shown as a ?ve-year moving average). As regards Europe, the story told by Figure VIII is consistent with the characterization of the Bretton Woods system as a period of when true exchange rate stability was remarkably short-lived. From 1946 until the arrival of the late 1950s, while Europe was not oating in the modern sense—as most currencies were not 31. We plot the premium rather than the market-determined rate, as it allows us to aggregate across countries in comparable units (percent). TABLE VII FLOATING PEGS AND PEGGED FLOATS: REVISITING THE PAST, 1970–2001 Conditional probability that the regime is: In percent “Other” according to NC a conditional on being classi?ed “Peg” by IMF 44.5 “Peg” or “Limited Flexibility” according to NC conditional on being classi?ed “Managed Floating” by IMF 53.2 “Peg” or “Limited Flexibility” according to NC conditional on being classi?ed “Independently Floating” by IMF 31.5 Pairwise correlation between IMF and NC classi?cations 42.0 Sources: The authors’ calculations. a. NC refers to the Natural Classi?cation; “Other” according to NC includes limited exibility, managed oating, freely oating, and freely falling. EXCHANGE RATE ARRANGEMENTS 33FIGURE VIII Average Monthly Parallel Market Premium: 1946–1998 Sources: International Monetary Fund, Annual Report on Exchange Arrangements and Exchange Restrictions and International Financial Statistics; Pick and Se´dillot [1971]; International Currency Analysis, World Currency Yearbook, various issues. The solid line represents the average monthly parallel market premium while the dashed line shows the ?ve-year moving average of plus one standard deviation. The regional averages are calculated excluding the freely falling episodes. 34 QUARTERLY JOURNAL OF ECONOMICSconvertible—it had some variant of de facto oating under the guise of pegged of?cial exchange rates. Each time of?cial rates are realigned, the story had already unfolded in the parallel market (as shown earlier in Figure II). While the volatility of the gap between the of?cial rate and the market exchange rate is not quite in the order of magnitude observed in the developing world, the volatility of the parallel rate is quite similar to the volatility of today’s managed or freely oating exchange rates. 32 There are many cases that illustrate clearly that little changed before and after the breakup of Bretton Woods. 33 Clearly, more careful statistical testing is required to make categorical statements about when a structural break took place; but it is obvious from the ?gures that whatever break might have taken place hardly lives up to the usual image of the move from ?xed to exible rates. IV.E. The Floats That Peg Figure IX provides a general avor of how exchange rate exibility has evolved over time and across regions. The ?gure plots ?ve-year moving averages of the probability that the monthly percent change in the exchange rate remains within a 2 percent band for Africa, Asia, Europe, and Western Hemisphere (excluding only the United States). Hence, under a pegged arrangement, assuming no adjustments to the parity, these probabilities should equal 100 percent. As before, we exclude the freely falling episodes. For comparison purposes, the ?gures plot the unweighted regional averages against the unweighted averages for the “committed oaters.” (The committed oaters include the following exchange rates against the dollar: Yen, DM (euro), Australian dollar, and the UK pound.) The dashed lines, which show plus/minus one standard deviation around the regional averages, highlight the differences between the group of oaters and the regional averages. It is evident for all regions (this applies the least to Africa) that the monthly percent variation in the exchange rate has 32. See Bordo [1993] on Bretton Woods and Bordo [2003] on a historical perspective on the evolution of exchange rate arrangements. 33. The country-by-country ?gures in “The Country Chronologies and Chartbook, Background Material to A Modern History of Exchange Rate Arrangements: A Reinterpretation” at http://www.puaf.umd.edu/faculty/papers/reinhart/ reinhart.htm are particularly revealing in this regard. EXCHANGE RATE ARRANGEMENTS 35FIGURE IX Absolute Monthly Percent Change in the Exchange Rate: Percent of Observations within a 62 Percent Band (?ve-year moving average) Sources: International Monetary Fund, Annual Report on Exchange Arrangements and Exchange Restrictions and International Financial Statistics; Pick and Se´dillot [1971]; International Currency Analysis, World Currency Yearbook, various issues. The solid line represents the average for the group while the dashed lines show plus/minus one standard deviation. The regional averages are calculated excluding the freely falling episodes. 36 QUARTERLY JOURNAL OF ECONOMICStypically been kept to a minimum—there is a great deal of smoothing of exchange rate uctuations in all regions when compared with the usual monthly variations of the committed oaters. The smoothing is most evident in Asia where the index hovers around 90 percent for most of the period, versus 60 –70 percent for the oaters. Hence, over time, the nature of the classi?cation problem has evolved from labeling something as a peg when it is not, to labeling something as oating when the degree of exchange rate exibility has in fact been very limited. IV.F. Does the Exchange Rate Regime Matter? The question of whether the exchange rate arrangement matters for various facets of economic activity has, indeed, been a far-reaching issue over the years in the literature on international trade and ?nance, and is beyond the scope of this paper. In this subsection we present a few simple exercises that do not speak to possible causal patterns between exchange rate regimes and economic performance, but are meant as illustrative of the potential usefulness of our classi?cation. First, consider Table VIII, which separates dual/parallel markets from all the other regimes where the “exchange rate is unitary,” to employ the language of the IMF. The top row shows average ination rates and real per capita GDP growth for the period 1970 –2001 for dual arrangements separately from all other regimes. This two-way split drastically alters the picture presented by the IMF’s classi- ?cation in the top and fourth rows of Table IX, which does not TABLE VIII INFLATION AND PER CAPITA REAL GDP GROWTH: A COMPARISON OF DUAL (OR MULTIPLE) AND UNIFIED EXCHANGE RATE SYSTEMS, 1970–2001 Regime Average annual ination rate Average per capita real GDP growth Uni?ed exchange rate 19.8 1.8 Dual (or multiple) exchange rates 162.5 0.8 Sources: International Monetary Fund, Annual Report on Exchange Arrangements and Exchange Restrictions and International Financial Statistics, Pick and Se´dillot [1971], International Currency Analysis, World Currency Yearbook, various issues. The averages for the two regime types (uni?ed and dual) are calculated on a country-by-country and year-by-year basis. Thus, if a country has a uni?ed exchange rate for most of the year, the observation for that year is included in the averages for uni?ed rates; if in the following year that same country introduces a dual market (or multiple rate) for most of the year, the observation for that year is included in the average for dual rates. This treatment allows us to deal with transitions across regime types over time. EXCHANGE RATE ARRANGEMENTS 37treat dual markets as a separate category. Dual (or multiple) exchange rate episodes are associated with an average ination rate of 163 percent versus 20 percent for uni?ed exchange markets—growth is one percentage point lower for dual arrangements. The explanation for this gap between the outcomes shown in Table VIII and the IMF’s in Table IX is twofold. First, 62 percent of the freely falling cases during 1970 –2001 were associated with parallel markets or dual or multiple exchange rates. Second, the high ination cases classi?ed by the IMF as freely oating were moved to the freely falling category Natural classi- ?cation. Again, we caution against overinterpreting the results in Table VIII as evidence of causality, as exchange controls and dual markets are often introduced amid political and economic crises—as the recent controls in Argentina (2001) and Venezuela (2003) attest. As Table IX highlights, according to the IMF, only limited exibility cases record moderate ination. On the other hand, freely oating cases record the best ination performance (9 percent) in the Natural classi?cation. Freely falling regimes exhibit an average annual ination rate 443 percent versus an ination average in the 9 to 17 percent range for the other categories (Table IX). TABLE IX DO CLASSIFICATIONS MATTER? GROWTH, INFLATION, AND TRADE ACROSS REGIMES: 1970–2001 Classi?cation scheme Peg Limited exibility Managed oating Freely oating Freely falling Average annual ination rate IMF Of?cial 38.8 5.3 74.8 173.9 n.a. Natural 15.9 10.1 16.5 9.4 443.3 Average annual per capita real GDP growth IMF Of?cial 1.4 2.2 1.9 0.5 n.a. Natural 1.9 2.4 1.6 2.3 22.5 Exports plus imports as a percent of GDP IMF Of?cial 69.9 81.0 65.8 60.6 n.a. Natural 78.7 80.3 61.2 44.9 57.1 Source: International Monetary Fund, World Economic Outlook. An n.a. denotes not available. The averages for each regime type (peg, limited exibility, etc.) are calculated on a country-by-country and year-by-year basis. Thus, if a country has a pegged exchange rate for most of the year, the observation for that year is included in the averages for pegs; if in the following year that same country has a managed oat for most of the year, the observation for that year is included in the average for managed oats. This treatment allows us to deal with transitions across regime types over time. 38 QUARTERLY JOURNAL OF ECONOMICSThe contrast is also signi?cant both in terms of the level of per capita GDP (Figure X) and per capita growth (Figure XI and Table IX). Freely falling has the lowest per capita income (US $3,476) of any category—highlighting that the earlier parallel to the HIPC debtor is an apt one—while freely oating has the highest (US $13,602). In the of?cial IMF classi?cation, limited exibility, which was almost entirely comprised of European countries, shows the largest per capita income. Growth is negative for the freely falling cases (22.5 percent) versus growth rates in the 1.6 –2.4 percent range for the other categories. Once freely falling is a separate category, the differences between our other classi?cations pale relative to the differences between freely falling and all others (Table VIII). In the of?cial IMF classi?cation, freely oating shows a meager average growth rate of 0.5 percent for the independently oating cases. For the Natural classi?cation, the average growth rate quadruples for the oaters to 2.3 percent. Clearly, this exercise highlights the importance of treating the freely falling episodes separately. FIGURE X PPP Adjusted GDP per Capita across Regime Types: 1970–2001 (averaging over all regions) EXCHANGE RATE ARRANGEMENTS 39V. CONCLUDING REMARKS According to our Natural classi?cation, across all countries for 1970 –2001, 45 percent of the observations of?cially labeled as a “peg” should, in fact, have been classi?ed as limited exibility, managed or freely oating— or worse, “freely falling.” PostBretton Woods, a new type of misclassi?cation problem emerged, and the odds of being of?cially labeled a “managed oat” when there was a de facto peg or crawling peg were about 53 percent. We thus ?nd that the of?cial and other histories of exchange rate arrangements can be profoundly misleading, as a striking number of pegs are much better described as oats, and vice versa. These misclassi?cation problems may cloud our view of history along some basic dimensions. Using the IMF’s classi?cation FIGURE XI Real per Capita GDP Growth across Regime Types: 1970–2001 (averaging over all regions) Sources: International Monetary Fund, Annual Report on Exchange Arrangements and Exchange Restrictions and International Financial Statistics; Pick and Se´dillot [1971]; International Currency Analysis, World Currency Yearbook, various issues. The averages for each regime type (peg, limited exibility, etc.) are calculated on a country-by-country and year-by-year basis. Thus, if a country has a pegged exchange rate for most of the year, the observation for that year is included in the averages for pegs; if in the following year that same country has a managed oat for most of the year, the observation for that year is included in the average for managed oats. This treatment allows us to deal with transitions across regime types over time. 40 QUARTERLY JOURNAL OF ECONOMICSfor the period 1970 to 2001, for instance, one would conclude that a freely oating exchange rate is not a very attractive option—it produces an average annual ination rate of 174 percent and a paltry average per capita growth rate of 0.5 percent. This is the worst performance of any arrangement. Our classi?cation presents a very different picture: free oats deliver an average in- ation that is less than 10 percent (the lowest of any exchange rate arrangement), and an average per capita growth rate of 2.3 percent. Equally importantly, we ?nd that uni?ed exchange rate regimes vastly outperform dual or multiple exchange rate arrangements, although one cannot necessarily interpret these differences as causal. While we have focused in this paper on the exchange rate arrangement classi?cation issue, the country histories and data provided in this paper may well have consequences for theory and empirics going forward, especially the issue of accounting for dual an parallel markets. In her classic history of the IMF de Vries [1969] looked back at the early years of the Bretton Woods regime and noted: Multiple exchange rates were one of the ?rst problems that faced the Fund in 1946, and have probably been its most common problem in the ?eld of exchange rates. An impressive number and diversity of countries in the last twenty years have experimented with one form or another of what the Fund has called multiple currency practices, at least for a few if not most of their transactions . . . The problem of multiple rates, then, never seems entirely at an end. Thirty-four years have passed since this history was written, and multiple exchange rate practices are showing no signs of becoming passe´ . On December 2001 Argentina suspended convertibility and, in so doing, segmented the market for foreign exchange, while on February 7, 2003, Venezuela introduced strict new exchange controls—de facto creating a multiple exchange rate system. Some things never change. APPENDIX: THE DETAILS OF THE “NATURAL” CLASSIFICATION This appendix describes the details of our classi?cation algorithm, which is outlined in Section III of the paper. We concentrate on the description of the ?ne grid as shown in Table V. A. Exchange Rate Flexibility Indices and Probability Analysis Our judgment about the appropriate exchange rate classi?- cation is shaped importantly by the time-series of several meaEXCHANGE RATE ARRANGEMENTS 41sures of exchange rate variability, based on monthly observations and averaged over two-year and ?ve-year rolling windows. The ?rst of these measures is the absolute percent change in the monthly nominal exchange rate. We prefer the mean absolute change to the variance to minimize the impact of outliers. These outliers arise when, for example, there are long periods in which the exchange rate is ?xed but, nonetheless, subject to rare but large devaluations. To assess whether exchange rate changes are kept within a band, we calculate the probabilities that the exchange rate remains within a plus/minus 1, 2, and 5 percent-wide band over any given period. Two percent seems a reasonable cutoff to distinguish between the limited exibility cases and more exible arrangements, as even in the Exchange Rate Mechanism arrangement in Europe 62 1 4 bands were allowed. As with the mean absolute deviation, these probabilities are calculated over twoyear and ?ve-year rolling windows. Unless otherwise noted in the chronologies, we use the ?ve-year rolling windows as our primary measure for the reasons discussed in Section III of the paper. These rolling probabilities are especially useful to detect implicit unannounced pegs and bands. B. De Jure and de Facto Pegs and Bands Where the chronologies show the authorities explicitly announcing a peg, we shortcut the de facto dating scheme described below and zero in on the date announced as the start of the peg. We then con?rm (or not) the peg by examining the mean absolute monthly change over the period following the announcement. The chronologies we develop, which give the day, month, and year when a peg becomes operative, are essential to our algorithm. There are two circumstances where we need to go beyond simply verifying the announced peg. The ?rst case is where our chronologies indicate that the peg applies only to an of?cial rate and that there is an active parallel (of?cial or illegal) market. As shown in Figure III, in these cases we apply the same battery of tests to the parallel market exchange rate as we do to the of?cial rate in a uni?ed market. Second, there are the cases where the of?cial policy is a peg to an undisclosed basket of currencies. In these cases, we verify if the “basket” peg is really a de facto peg to a single dominant currency (or to the SDR). If no dominant currency can be identi?ed, we do not label the episode as a peg. Potentially, of course, 42 QUARTERLY JOURNAL OF ECONOMICSwe may be missing some de facto basket pegs, though in practice, this is almost certainly not a major issue. We now describe our approach toward detecting de facto pegs. If there is no of?cially announced peg, we test for a “de facto” peg in two ways. First, we examine the monthly absolute percent changes. If the absolute monthly percent change in the exchange rate is equal to zero for four consecutive months or more, that episode is classi?ed (for however long its lasts) as a de facto peg if there are no dual or multiple exchange rates. This allows us to identify short-lived de facto pegs as well as those with a longer duration. For instance, this ?lter allowed us to identify the Philippines’ de facto peg to the US dollar during 1995–1997 in the run-up to the Asian crisis as well as the numerous European de facto pegs to the DM well ahead of the introduction of the euro. Second, we compute the probability that the monthly exchange rate change remains within a 1 percent band over a rolling ?ve-year period: 34 P~e , 1%!, where e is the monthly absolute percentage change in the exchange rate. If this probability is 80 percent or higher, then the regime is classi?ed as a de facto peg or crawling peg over the entire ?ve-year period. If the exchange rate has no drift, it is classi?ed as a ?xed parity; if a positive drift is present, it is labeled a crawling peg; and, if the exchange rate also goes through periods of both appreciation and depreciation, it is dubbed a “noncrawling” peg. Our choice of an 80 percent threshold is not accidental, but rather we chose this value because it appears to do a very good job at detecting regimes one would want to label as pegs, without drawing in a signi?cant number of “false positives.” Our approach regarding preannounced and de facto bands follows exactly the same process as that of detecting preannounced and de facto pegs, we simply replace the 61% band with a 62% band in the algorithm. If a band is announced and the chronologies show a uni?ed exchange market, we label the episode as a band unless it had already been identi?ed as a de facto peg by the criteria described earlier. But, importantly, we also verify whether the announced and de facto bands coincide, espe- 34. There are a handful of cases where a two-year window is used. In such instances, it is noted in the chronologies. EXCHANGE RATE ARRANGEMENTS 43cially as there are numerous cases where the announced (de jure) band is much wider than the de facto band. 35 To detect such cases, we calculate the probability that the monthly exchange rate change remains within a 62% band over a rolling ?ve-year period: P~e , 2%!. If this probability is 80 percent or higher, then the regime is classi?ed as a de facto narrow horizontal, crawling, or noncrawling band (which allows for both a sustained appreciation and depreciation) over the period through which it remains continuously above the 80 percent threshold. In the case where the preannounced bands are wide (meaning equal to or greater than 65%), we also verify 65% bands. The speci?cs for each case are discussed in the country chronologies. For instance, as shown earlier in Table IV, in the case of Chile we found that the de facto band during 1992–1998 was narrower (65%) than that which was announced at the time (610% and 612.5%). In the case of Libya, which had an announced 77 percent wide band along a ?xed central parity pegged to the SDR over the March 1986 –December 2001, we detected a 65% crawling band to the US dollar. C. Freely Falling As we emphasize in the text, there are situations, almost invariably due to high ination or hyperination, in which there are mega-depreciations in the exchange rate on a routine and sustained basis. We have argued that it is inappropriate and misleading to lump these cases—which is what all previous classi?cations (IMF or otherwise) do—with oating rate regimes. We label episodes freely falling on the basis of two criteria. First, periods where the twelve-month rate of ination equals or exceeds 40 percent are classi?ed as freely falling unless they have been identi?ed as some form of preannounced peg or preannounced narrow band by the above criteria. 36 The 40 percent 35. Mexico’s exchange rate policy prior to the December 1994 crisis is one of numerous examples of this pattern. Despite the fact that the band was widening over time, as the oor of the band was ?xed and the ceiling was crawling, the peso remained virtually pegged to the US dollar for extended periods of time. 36. It is critical that the peg criteria supersede the high ination criteria in the classi?cation strategy, since historically a majority of ination stabilization efforts have used the exchange rate as the nominal anchor and in many of these episodes ination rates at the outset of the peg were well above our 40 percent threshold. 44 QUARTERLY JOURNAL OF ECONOMICSination threshold is not entirely arbitrary, as it has been identi?ed as an important benchmark in the literature on the determinants of growth (see Easterly [2001]). As a special subcategory of freely falling, we dub as hyperoats those episodes that meet Cagan’s [1956] classic de?nition of hyperination (50 percent or more ination per month). A second situation where we classify an exchange rate regime as freely falling are the six months immediately following a currency crisis—but only for those cases where the crisis marks a transition from a ?xed or quasi-?xed regime to a managed or independently oating regime. 37 Such episodes are typically characterized by exchange rate overshooting. This is another situation where a large change in the exchange rate does not owe to a deliberate policy; it is the reection of a loss of credibility and recurring speculative attacks. To date these crisis episodes, we follow a variant of the approach suggested by Frankel and Rose [1996]. Namely, any month where the depreciation exceeds or equals 12 1 2 percent and also exceeds the preceding month’s depreciation by at least 10 percent is identi?ed as a crisis. 38 To make sure that this approach yields plausible crisis dates, we supplement the analysis with our extensive country chronologies, which also shed light on balance of payments dif?culties. 39 Since, as a rule, freely falling is not typically an explicit arrangement of choice, our chronologies also provide for all the freely falling cases, the underlying de jure or de facto arrangement (for example, dual markets, independently oating, etc.). D. Managed and Freely Floating Our approach toward identifying managed and freely oating episodes is basically to create these classes out of the residual pool of episodes that, after comprehensive application of our algorithm, have not been identi?ed as an explicit or implicit peg or some form of band, and that are not included in the freely 37. This rules out cases where there was a devaluation and a repeg and cases where the large exchange rate swing occurred in the context of an already oating rate. 38. Frankel and Rose [1996] do not date the speci?c month of the crisis but the year; their criteria call for a 25 percent (or higher) depreciation over the year. 39. For instance, the Thai crisis of July 1997 does not meet the modi?ed Frankel-Rose criteria. While the depreciation in July exceeded that of the preceding month by more than 10 percent, the depreciation of the Thai Baht in that month did not exceed 25 percent. For these cases, we rely on the chronologies of events. EXCHANGE RATE ARRANGEMENTS 45falling category. To proxy the degree of exchange rate exibility under freely oating and managed oats, we construct a composite statistic, e/P~e , 1%!, where the numerator is the mean absolute monthly percent change in the exchange rate over a rolling ?ve-year period, while the denominator ags the likelihood of small changes. For de jure or de facto pegs, this index will be very low (close to or equal to zero), while for the freely falling cases it will be very large. As noted, we only focus on this index for those countries and periods which are candidates for freely or managed oating. We tabulate the frequency distribution of our index for the currencies that are most transparently oating, these include US dollar/DM-euro, US dollar/yen, US dollar/UK pound, US dollar/Australian dollar, and US dollar/New Zealand dollar beginning on the date in which the oat was announced. We pool the observations (the ratio for rolling ?ve-year averages) for all the oaters. So, for example, since Brazil oated the real in January 1999, we would calculate the ratio only from that date forward. If Brazil’s ratio falls inside the 99 percent con?dence interval (the null hypothesis is freely oating and hence the rejection region is located at the lower tail of the distribution of the oater’s group), the episode is characterized as freely oating. If that ratio falls in the lower 1 percent tail, the null hypothesis of freely oating is rejected in favor of the alternative hypothesis of managed oat. It is important to note that managed by this de?nition does not necessarily imply active or frequent foreign exchange market intervention—it refers to the fact that for whatever reason our composite exchange rate variability index, e/P(e , 1%), does not behave like the indices for the freely oaters. E. Dual or Multiple Exchange Rate Regimes and Parallel Markets Dual rates are essentially a hybrid arrangement. There are cases or periods in which the premium is nil and stable so that the of?cial rate is representative of the underlying monetary policy. The of?cial exchange rate could be pegged, crawling, or maintained within some bands, or in a few cases allowed to oat. But there are countless episodes where the divergence between the of?cial and parallel rate is so large that the picture is incomplete without knowledge of what the parallel market rate is doing. The 46 QUARTERLY JOURNAL OF ECONOMICScountry chronologies are critical in identifying these episodes. In the cases where dual or multiple rates are present or parallel markets are active, we focus on the market-determined rates instead of the of?cial exchange rates. As shown in Figure III, we subject the market-determined exchange rate (dual, multiple, or parallel) to the battery of tests described above. 40 This particular category will especially reshape how we view the 1940s through the 1960s, where about half the cases in the sample involved dual markets. UNIVERSITY OF MARYLAND, COLLEGE PARK HARVARD UNIVERSITY REFERENCES Bahmani-Oskooee, Mohsen, Ilir Miteza, and A. B. M. Nasir, “The Long-Run Relationship between Black Market and Of?cial Exchange Rates: Evidence from Panel Cointegration,” Economics Letters, LXXVI (2002), 397–404. Baxter, Marianne, and Alan Stockman, “Business Cycle and Exchange Rate Regime: Some International Evidence,” Journal of Monetary Economics, XXIII (1989), 377– 400. Bordo, Michael, “The Bretton Woods International Monetary System: A Historical Overview,” in A Retrospective on the Bretton Woods System, Michael Bordo and Barry Eichengreen, eds. (Chicago, IL: University of Chicago Press, 1993), pp. 3–98. ——, “Exchange Rate Regimes in Historical Perspective,” National Bureau of Economic Research Working Paper No. 9654, 2003. Cagan, Philip, “The Monetary Dynamics of Hyperination,” in Studies in the Quantity Theory of Money, Milton Friedman, ed. (Chicago, IL: University of Chicago Press, 1956), pp. 25–117. Calvo, Guillermo A., and Carmen M. Reinhart, “Fear of Floating,” Quarterly Journal of Economics, CXVII (2002), 379– 408. Chen, Yu-chin, and Kenneth S. Rogoff, “Commodity Currencies,” Journal of International Economics, VX (2003), 133–160. Claessens, Stijn, “Estimates of Capital Flight and Its Behavior,” Revista de Ana´ lisis Econo´mico, XII (1997), 3–34. Cotarelli, C., and C. Giannini. “Credibility Without Rules? Monetary Frameworks in the Post Bretton-Woods Era,” IMF Occasional Paper No. 154 (Washington, DC: International Monetary Fund, 1998). de Vries, Margaret G., “Multiple Exchange Rates,” in The International Monetary Fund 1945–1965, Margaret de Vries and J. Keith Horse?eld, eds. (Washington, DC: International Monetary Fund, 1969), pp. 122–151. Easterly, William, The Elusive Quest for Growth (Cambridge, MA: MIT Press, 2001). Frankel, Jeffrey A., and Andrew K. Rose, “Currency Crashes in Emerging Markets: An Empirical Treatment,” Journal of International Economics, XXXXI (1996), 351–368. Ghei, Nita, Miguel A. Kiguel, and Stephen A. O’Connell, “Parallel Exchange Rates in Developing Countries: Lessons from Eight Case Studies,” in Parallel Exchange Rates in Developing Countries, Miguel Kiguel, J. Saul Lizondo, and Stephen O’Connell, eds. (New York, NY: Saint Martin’s Press, 1997), pp. 17–76. 40. There are a few such cases in the sample, where only government transactions take place at the of?cial rate. EXCHANGE RATE ARRANGEMENTS 47Ghosh, Atish, Anne-Marie Gulde, Jonathan Ostry, and Holger Wolfe, “Does the Nominal Exchange Rate Regime Matter?” National Bureau of Economic Research Working Paper No. 5874, 1997. International Currency Analysis, World Currency Yearbook (New York, NY: International Currency Analysis, 1983–1998), various issues. International Monetary Fund, Annual Report on Exchange Restrictions (Washington, DC: International Monetary Fund, 1949–1978), various issues. International Monetary Fund, Annual Report on Exchange Arrangements and Exchange Restriction (Washington, DC: International Monetary Fund, 1979– 2001), various issues. Kiguel, Miguel, J. Saul Lizondo, and Stephen A. O’Connell, eds., Parallel Exchange Rates in Developing Countries (New York, NY: Saint Martin’s Press, 1997). Levy-Yeyati, Eduardo, and Federico Sturzenegger, “Classifying Exchange Rate Regimes: Deeds versus Words,” mimeo, Universidad Torcuato Di Tella, 2002. Marion, Nancy P., “Dual Exchange Rates in Europe and Latin America,” World Bank Economic Review, VIII (1994), 213–245. Pick, Franz, World Currency Reports (New York, NY: Pick Publishing Corporation, 1945–1955), various issues. ——, Black Market Yearbook (New York, NY: Pick Publishing Corporation, 1951– 1955), various issues. ——, Pick’s Currency Yearbook (New York, NY: Pick Publishing Corporation, 1955–1982), various issues. ——, World Currency Reports (New York, NY: International Currency Analysis Inc., 1983–1998), various issues. Pick, Franz, and Rene´ Se´dillot, All the Monies of the World: A Chronicle of Currency Values (New York, NY: Pick Publishing Corporation, 1971). Reinhart, Carmen M., and Kenneth S. Rogoff, “A Modern History of Exchange Rate Arrangements: A Reinterpretation,” National Bureau of Economic Research Working Paper No. 8963, 2001. Reinhart, Carmen M., and Kenneth S. Rogoff, “Parts I and II. Background Material to a Modern History of Exchange Rate Arrangements: A Reinterpretation,” mimeo, International Monetary Fund, Washington, DC, 2003 at http:// www.puaf.umd.edu/faculty/papers/reinhart/reinhart.htm. Reinhart, Carmen M., Kenneth S. Rogoff, and Miguel A. Savastano, “Addicted to Dollars,” National Bureau of Economic Research Working Paper No. 10015. 2003. Reinhart, Carmen M., Kenneth S. Rogoff, and Antonio Spilimbergo, “When Hard Shocks Hit Soft Pegs,” mimeo, International Monetary Fund, Washington, DC, 2003. United Nations, United Nations Yearbook (New York: United Nations, 1946– 1960), various issues. 48 QUARTERLY JOURNAL OF ECONOMICS