The effect of the exchange rate regime on inflation and growth is examined. The 30-year data set includes over 100 countries and nine regime types. Pegged regimes are associated with lower inflation than intermediate or flexible regimes. This anti-inflationary benefit reflects lower money supply growth (a discipline effect) and higher money demand growth (a credibility effect). Output growth does not vary significantly across regimes: Countries with pegged regimes invest more and are more open to international trade than those with flexible rates, but they experience lower residual productivity growth. Output and employment are more variable under pegged rates than under flexible rates.

Abstract

The effect of the exchange rate regime on inflation and growth is examined. The 30-year data set includes over 100 countries and nine regime types. Pegged regimes are associated with lower inflation than intermediate or flexible regimes. This anti-inflationary benefit reflects lower money supply growth (a discipline effect) and higher money demand growth (a credibility effect). Output growth does not vary significantly across regimes: Countries with pegged regimes invest more and are more open to international trade than those with flexible rates, but they experience lower residual productivity growth. Output and employment are more variable under pegged rates than under flexible rates.

I. Introduction

Does the choice of the nominal exchange rate regime matter for macroeconomic performance? A lively theoretical debate has yet to yield unambiguous answers: increased nominal exchange rate flexibility has been argued both to aggravate and to reduce output variability, to enhance and to suppress trade, to raise and to lower investment, to foster and to reduce fiscal discipline, to increase and to decrease inflation. 2/ The quest for general results is impeded by the multitude of partly offsetting and partly reinforcing potential linkages between the regime and the key variables. A resolution of the theoretical debate will be difficult until these linkages can be ranked in terms of their importance—an empirical task we undertake in this paper.

Our results are based on an annual classification of the exchange rate system of up to 136 countries over the period 1960-1990 into nine regimes, covering single currency pegs, secret and published basket pegs, cooperative systems, crawling pegs, target zones, and floats with heavy, light, or no intervention. While–in line with most of the literature–we also report results for the less disaggregated classification of regimes into “pegged,” “intermediate” and “floating,” this three-way classification—and more so the traditional dichotomy between fixed and floating rates—loses much of the richness of real world regimes captured by our nine-way classification. To take just one example; while we generally find inflation to be positively related to the flexibility of the exchange rate regime, inflation under pure floats turns out to be—for the group of industrialized countries—actually lower than inflation under the dirty float regimes.

Our empirical analysis is divided into three parts. We begin by reporting the unconditional means of inflation and growth across regimes. Next, we evaluate the importance of alternative channels of interaction between the regime and macroeconomic performance by examining the residual effect of the regime on inflation and growth conditional on controlling for a set of other potential determinants. In a third stage, we explicitly allow for the possibility of reverse causation: a positive correlation between the flexibility of the exchange rate regime and average inflation could be a reflection of reduced monetary discipline in the absence of a peg, but it could also simply be the result of countries experiencing higher average inflation rates—for whatever reason—being more likely to choose flexible exchange rate regimes. In either case, we would observe a link between fixed exchange rate regimes and low inflation. Yet in the first case, the choice of regime is instrumental in determining macroeconomic performance, in the latter it is merely incidental. We address this problem by allowing for endogenous regime choice in a simultaneous equation framework.

Our main results on inflation, growth, and the business cycle may be summarized briefly. While there are important differences among the various forms of pegged exchange rate regimes, we find—with only a few exceptions—that inflation has generally been lower under pegged regimes than under more flexible arrangements. This result stems from two factors. First, a monetary discipline effect: fixed exchange rate regimes are associated with slower rates of monetary growth. Second, a confidence effect: fixed exchange rate regimes are associated with slower velocity growth (and hence higher money demand growth), thus yielding a lower inflation rate for a given rate of monetary expansion.

As regards growth, we find little systematic difference in performance across regimes. The factors driving growth are, however, quite different. Simple growth accounting suggests two ways in which the exchange rate regime could be related to observed differences in GDP growth: factor accumulation and productivity growth. We find that investment indeed differs systematically across nominal exchange rate regimes, being significantly higher under fixed than under intermediate regimes; and also higher under intermediate regimes than under floating exchange rate regimes. We split productivity growth differences into a part due to differences in trade performance—often argued to be affected by the exchange rate regime—and a residual component. Trade growth is found to have been significantly faster under fixed exchange rate regimes, while residual productivity growth is very much larger under flexible rates, and indeed sufficiently so to offset the growth effects of higher trade growth and higher investment under fixed rates.

Finally, we turn from the average growth rates of prices and output to their volatility, finding modest evidence that GDP growth is more volatile under pegged exchange rate regimes, and considerable evidence that employment volatility is increasing in the degree of exchange rate rigidity.

The paper is divided into four parts. Section II describes the data set. Our results on the link between the exchange rate regime and inflation and growth are reported in sections IIIandIV. Section V concludes.

II. Data

The customary distinction between “fixed” and “flexible” exchange rates does little justice to the rich variety of real world exchange rate systems, which span a continuum from the classic single-currency peg to basket pegs, cooperative agreements, target zones, crawling pegs and dirty floats all the way to pure floats. We opt for a fairly disaggregated classification. In classifying countries, two approaches are available. The first classifies exchange rate regimes according to the actual volatility of the nominal exchange rate. The approach has obvious appeal in being based on observable behavior, but it fails to capture the degree of commitment of the central bank to intervening in, and subordinating its monetary policy to, the foreign exchange market. The drawback can, in principle, be partly overcome by including both the exchange rate and policy variables, notably intervention, in the classification scheme. Yet even if augmented in this way, this performance-based approach is unable to distinguish between low volatility of the exchange rate due to an activist policy and low volatility due to lower volatility in the underlying disturbances.

The alternative approach—which we adopt in this paper—classifies regimes according to the stated intention of the central bank regarding its intervention policy, as summarized by the International Monetary Fund’s Annual Report on Exchange Arrangements and Exchange Restrictions. This approach is not without drawbacks, either. In particular, a country officially on a fixed exchange rate standard may adjust the peg so frequently as to transform a de jure peg into a de facto float. In our empirical work we address this problem by checking the robustness of our results against a secondary classification dividing the nonfloating regimes into frequent and infrequent adjusters. 3/

The optimal degree of disaggregation involves a trade-off between balanced statistical design and the risk of obscuring results by combining dissimilar categories. Our first classification, to which much of the analysis in the text will refer, divides exchange rate regimes into three broad categories. The first consists of pegged exchange rate regimes: single currency or basket pegs. The second comprises intermediate exchange rate regimes that fall between pegging and floating; this category includes crawling pegs and target zones. The third category consists of floating regimes.

Where relevant, we expand this three way taxonomy into nine categories distinguishing between types of exchange rate peg (single or multiple currency, published or secret basket), cooperative systems, and varieties of floating regimes. We use this broader classification to examine within category effects, for example, whether the “credibility” effect of pegging differs between a published and a secret basket peg. Also, when relevant to the results, we differentiate between the de jure nature of the regime and its de facto implementation by dividing the pegged and intermediate regimes into frequent and infrequent adjusters of the peg. This last classification is based on a survey of IMF desk officers.

The macroeconomic data are from the International Monetary Fund’s International Financial Statistics and World Economic Outlook databases. The original data set covers 136 countries over the period 1960-1990, yielding 3685 nonmissing observations for inflation, and 3732 observations on GDP growth. For the econometric analysis, a number of additional variables were added, including broad money growth rates, interest rates, the terms of trade, a variety of national accounts data (government consumption, investment, exports and imports) and measures of central bank independence, used mainly as instruments for the nominal exchange rate regime. The latter are taken from Cukierman (1992) who provides a detailed explanation of them. We use his measure of the turnover rate of the central bank governor, and three variables intended to measure the legal aspects of central bank independence, relating to the appointment of the governor; monetary policy formulation, and lending limits of the central bank. While the inclusion of the explanatory variables narrows the available sample significantly, most of the eliminated observations date from the 1960s. 4/ As the overwhelming preponderance of the single currency peg in the 1960s, in any case limits the scope for comparisons across regimes in this period, the loss of information is more modest than the loss of data would suggest.

Finally, for most of the tables we report results disaggregated by the income level of the country, according to the World Bank classification of countries into an industrialized and upper-middle income and a lower-middle income and low-income group. In addition, we group countries by the degree of capital controls, based on the International Monetary Fund’s Exchange Arrangements and Exchange Restrictions.

III. Inflation

A substantial body of literature predicts a positive correlation between exchange rate flexibility and inflation. 5/ Two reasons are typically cited. First, by providing a highly visible commitment, the adoption of a pegged exchange rate regime may raise the political costs of excessive monetary growth and the attendant collapse of the peg. Second, to the degree that the peg is credible, the growth of money demand may be more robust, thus reducing the inflationary consequences of a given monetary growth rate.

We examine the link between the exchange rate regime and inflation in three steps. First, we report the average inflation rate for the various regime classifications and country groupings. Second, we examine whether the regime exerts an effect controlling for other determinants of inflation. Finally, we allow for endogeneity of the exchange rate regime.

1. Unconditional means

Over our entire sample of 3685 observations, inflation (π) has averaged 13.7 percent per year: 11.5 percent in countries that had some form of exchange rate peg, 21.5 percent in countries that followed one of the intermediate exchange rate regimes, and 24.2 percent in countries that allowed their currency to float. Separating the nonfloating regimes by the frequency of adjustments of the peg reveals the importance of de facto behavior: the frequent adjusters experienced an average inflation rate of 25.2 percent as compared to 11.1 percent for regimes that entailed no or infrequent changes in the parity.

To some degree, these differences may reflect the fact that fixed regime observations are heavily concentrated in the 1960s, while the flexible regime observations are bunched in the 1970s and 1980s. Without judging whether the incidence of shocks and the trend increase in inflation over the sample is exogenous—or itself a function of exchange rate regime choice (an issue taken up below)—we control for this possibility by computing for each regime type, the average of the deviation of the inflation rates from the annual average of all observations. 6/ Table 1 reports the results, both for the entire group of countries and separately for countries in the high/upper-middle income and low/lower-middle income grouping used by the World Bank, and for countries without capital controls (using the IMF classification). To control for possible distortions from outliers, Table 1 also reports the means of π/(1+π), which is normalized between 0 and 1.

Table 1:

Average Inflation Rate

(Deviation From Annual Global Mean)

article image

Unconstrained mean

Normalized (π/1+π)

The first three rows contain the means for our first classification, revealing that controlling for the annual global average inflation does not materially affect the ranking: countries operating under pegged rates experienced below average inflation rates while countries under floating rates suffered inflation 7 percent above the average of all countries. The second three rows, containing the results for our third, performance-based, classification, again strongly suggests that what matters is the de facto, not the de jure regime: pegged regimes with frequent adjustments suffered above average inflation rates. The last nine rows report the results for the disaggregated classification, revealing the first instance of an interesting and—as we will see below—quite robust non-linearity: while increasing flexibility is generally associated with higher inflation rates, upper-income countries adopting the least restrictive system, the pure float, enjoy below average inflation, while low-income countries under pure floats suffer the highest average inflation of any regime. As neither of these rankings is affected by using the normalized inflation instead, the results are not driven by outliers. We hence proceed in the remainder of the paper using the simple inflation rates.

2. Conditional means

We next turn to examining the alternative channels linking inflation performance to the exchange rate regime, using the money market equilibrium condition as a reference point for choosing the conditioning variables:

MtPt=YtaIt-βVt(1)

where M and P denote money and the price level and V measures residual velocity controlling for income Y and interest rate I effects. It bears emphasizing that this money demand function is used simply to provide an interpretative framework; none of the results depend on the dozen money demand function itself. Taking logs (denoted by lower case letters), time-differentiating and re-arranging yields:

Π=Δpc=Δmc-αΔyc+βΔic+Δυc(2)

Equation (2) suggests four potential sources of differences in inflation across regimes: differences in monetary growth rates, differences in interest rate growth, differences in output growth and, residually, differences in velocity growth not accounted for by output and interest rate differentials. As our results in the next section suggest that aggregate output growth, in fact, does not differ significantly across regimes, we focus here on the remaining three potential causes of differences in inflation, beginning with OLS regressions for the full dataset before allowing for endogeneity of regime choice in a smaller data set. 7/

a. OLS results

Monetary growth rates differed significantly across regimes, averaging 16 percent under pegged regimes, 22 percent under intermediate regime, and 25 percent under flexible rate regimes. Again, these differences may partly reflect an accident of timing. The middle column of Table 2 reports the average of the difference between the monetary growth rates under a particular regime and the annual average for all countries. As before, removing world means does not materially affect the result: monetary growth under pegged rates was 7.5 percentage points lower than under floating rates. Turning to the more detailed decomposition reveals that while the negative association between monetary growth rates and exchange rate rigidity is common across groups, countries with open capital markets were characterized almost uniformly by lower monetary growth rates. Furthermore, the dichotomy of the pure floaters again resurfaces: while developing countries on pure floats experienced above average monetary growth rates, industrialized pure floaters experienced a significantly below-average monetary growth rate.

Table 2:

Inflation Regressions

article image
article image
* (**. ***) denotes significance at the 10 (5,1) percent level.

To assess the effect of the exchange rate regime, we include an exchange rate regime dummy Peg, equal to one if the exchange rate is pegged and zero otherwise, in a regression of the inflation rate on a constant, annual dummies, output growth (Δy), the turnover of the central bank governor (Turn) and openness (Open). 8/ The turnover variable was found by Cukierman (1992) to be the single most important determinant of inflation in his study of various central bank independence proxies. The openness variable is included to proxy a variety of effects including the higher costs of monetary expansion in open economies [Romer (1993), Lane (1994)] and the strength of international arbitrage constraints. Notice that this regression does not control for either the monetary growth rate nor the interest rate growth, so that, to the extent that these variables are connected with the regime, the effect will be captured by the exchange rate dummy, peg. The estimated coefficients are reported under the heading “Regression (1)” in Table 2.

For the whole sample, the coefficient on Peg is -0.059, so that inflation is on average almost 6 percentage points lower under pegged rates than under other regimes.

Conceptually, this 6 percentage point difference reflects three factors: a monetary discipline effect; a money demand effect which would be manifest through a faster decline in interest rates; and a residual confidence of the pegged exchange rate regime. Moving beyond the regression reported in column (1), therefore, it is instructive to examine whether there is any residual confidence effect after controlling for growth rates for money, (Δm), and of interest rates, (Δi). 9/ The regression becomes: 10/

Π=-0.748Δy+0.089Turn-0.003Open+0.852Δm+0.015Δi-0.014Peg(6.84***)(2.62**)(4.48***)(20.19***)(0.82)(2.00*)R2=0.86

interest rate growth, so that, to the extent that these variables are connected with the regime, the effect will be captured by the exchange rate dummy, peg. The estimated coefficients are reported under the heading “Regression (1)” in Table 2.

For the whole sample, the coefficient on Peg is -0.059, so that inflation is on average almost 6 percentage points lower under pegged rates than under other regimes.

Conceptually, this 6 percentage point difference reflects three factors: a monetary discipline effect; a money demand effect which would be manifest through a faster decline in interest rates; and a residual confidence of the pegged exchange rate regime. Moving beyond the regression reported in column (1), therefore, it is instructive to examine whether there is any residual confidence effect after controlling for growth rates for money, (Δm), and of interest rates, (Δi). 11/ The regression becomes: 12/

Π=-0.748Δy+0.089Turn-0.003Open+0.852Δm+0.015Δi-0.014Peg(6.84***)(2.62**)(4.48***)(20.19***)(0.82)(2.00*)R2=0.86

where numbers in brackets are t-statistics based on White standard errors and one, two and three stars denote significance at the 10, 5 and 1 percent level. The regression was replicated for each of the detailed classifications of the regime and separately by income group and by the degree of capital mobility. The estimated coefficient on the Peg indicator variable, together with the associated t-statistic and the R2 of the regression, is reported under the heading “Regression (2)” in Table 2.

Higher output growth and higher openness reduces inflation, while a higher turnover of the central bank governor increases inflation ceteris paribus. Monetary growth enters highly significant, e.g., with a coefficient near unity. Interest rate growth enters with the expected sign but is insignificant.

The coefficient on peg now falls to -0.014; that is the residual confidence effect of a pegged exchange rate regime lowers inflation by 1.4 percentage points, even controlling for the greater monetary discipline and the faster decline of interest rates associated with pegged regimes. This residual confidence effect, which is both economically and statistically significant, means that pegging the exchange rate brings additional anti inflationary benefits beyond discipline and the standard determinants of velocity.

The detailed results suggest that this confidence effect is strongest for the single currency peg and, surprisingly, the secret rather than the published basket pegs. 13/ As before, the distinction between frequently and infrequently adjusted pegs suggests that it is the de facto rather than the de jure regime which matters [Svensson (1993)]: while the infrequent adjusters had inflation that was 13 percentage points below the average in the sample, the inflation benefit for frequent adjusters drops to 3 percent. The asymmetry of the pure float mentioned above is again present: for the high-income countries inflation under the pure float was below average, while low-income countries on pure floats experienced an inflation rate almost 5 percentage points above the average.

b. Endogeneity of the exchange rate regime

The results presented above—a significant positive correlation between exchange rate flexibility and inflation—permit two (nonexclusive) interpretations: a causal link from an exogenous exchange rate regime to macroeconomic performance, or an endogenous regime choice of countries conditional on their macroeconomic performance. The distinction is of prime importance: in the first case, the choice of regime has important implications for macroeconomic performance, in the second, it is incidental. To be specific, suppose that the choice whether or not to peg depends on some set of variables X2 as well as on the inflation rate II:

Peg*=X2β2+γ2Π+η2(3)

where Peg* is an unobserved “desire” to peg the exchange rate and η captures non-systematic factors. Let Peg denote the observed indicator variable designating whether the country in fact has a pegged exchange rate, with Peg=1 if Peg* is above some critical value and 0 otherwise. A natural assumption might be γ2 < 0: low inflation countries are best able to maintain a pegged exchange rate regime and perhaps are more likely to want to do so. The structural equation determining inflation is given by:

Π-X1β1+1ϒpeg*+η1(4)

where X1 and η denote a vector of exogenous variables and an error term, and γ1 < 0. The simultaneity implies that the anti-inflationary benefit of pegged exchange rates identified in the previous section may be spurious: we may find a statistically significant negative estimate for γ1 even though its true value is zero. In order to address the possibility of this bias, we use a simultaneous equation framework. Finding adequate instruments for the exchange rate regime is, of course, no mean task. Here we use the variables defining the legal independence of the central bank, taken from Cukierman [1992]. It seems reasonable to assume that the factors that lead a country to adopt a particular stance in regard to the independence of its central bank might also influence its choice of an exchange rate (underline) regime. Cukierman finds that these variables tend not to be correlated with the inflation rate, and they are assumed not to enter the inflation equation (4).14/

As the endogenous variable is dichotomous, standard two stage estimation is not feasible. We, therefore, use a modification of Amemiya’s [1979] 2SLS method for truncated endogenous variables. The modification, proposed by Maddala [1983], explicitly allows for dichotomous variables. The reduced forms of the structural model are given by:

Π-Xλ1+υ1(5)
Peg*-Xλ2+ν2(6)

where X includes both X1 and X2 Since Peg* is only observed as a dichotomous variable, we can only estimate λ2*λ2σ2(υ2).

Defining Peg**=Xλ2*+υ2σ2(υ2), the structural inflation equation can then be rewritten as:

Π-X1β1+γ1σν22Peg**+(7)

where ∊ is an error term. The two-stage procedure then involves estimating λ*2 by probate maximum likelihood, calculating Peg** and substituting it into equation 7 which can then be estimated by OLS. If the adoption of a pegged exchange rate regime, in fact, exerts a negative effect on inflation, the estimated coefficient γ1συ22 γ1 σ2υ2 expected to be negative. 15/

For the second stage probate regression we use the three legal definitions of central bank independence developed by Cukierman [1992], 16/ the openness measure and the fitted value of inflation. 17/ Of these, the independence variable measuring the legal status of the central bank governor is significantly negative (that is, lowers the likelihood of a pegged exchange rate regime being chosen), while openness has a positive, and inflation the expected negative significant effect. Sixty-eight percent of all observations were correctly predicted. Allowing for regime choice endogeneity yields the following corrected OLS regression:

Π=-0.0054Open+0.624Δy+0.0285Turn+0.828Δm+0.017Δi-0.028Peg(5.91***)(7.56***)(0.83)(39.77***)(1.44)(2.06**)R2=0.84

The corresponding estimates for the intermediate and flexible regimes are 0.022 (t=0.69), and 0.018 (t-1.58). 18/ The results are comparable to those obtained above for the OLS regressions: the residual effect on velocity growth is again both economically and statistically significant. Allowing for endogenous regime choice, thus, does not materially affect the results for this small dataset, enhancing our confidence in the robustness of the OLS results reported in the previous section.

3. Volatility

We next examine how the volatility of inflation differs across alternative exchange rate regimes. To capture the notion that welfare costs primarily arise from unanticipated inflation, we use a three-year centered moving standard deviation of the residual from a regression of the inflation rate on its own lag as our inflation volatility measure. 19/ The first column of Table 3 reports average volatility across exchange rate regimes. As before, the entries refer to the deviation from the annual world average. The rankings are comparable with those obtained for the mean inflation rates, conforming to the familiar positive correlation between the level and volatility of inflation (Ball (1992)).

Table 3.

Inflation and Reserve Volatility Regression

article image
article image

Turning to the subsamples, the developing countries exhibited the greatest volatility of inflation rates regardless of the exchange rate regime, though the positive correlation between exchange rate flexibility and inflation volatility holds within each subgroup. The sharp split of results for the pure float between the upper and lower half of the income distribution is seen to extend to the volatility measure.

Again, however, the exchange rate regime is likely to be only one of many determinants of inflation volatility. To assess the conditional effect of the regime, we regress the measure of inflation volatility on output volatility, the central banker turnover rate, openness, and the volatility of money supply growth and interest rates, yielding:

σΠ=-0.242σ(Δy)+0.274Turn-0.0002Open+0.345σ(Δy)-0.00003σ(Δi)-0.0199Peg(1.04)(3.88***)(3.01**)(2.84**)(3.32***)(1.53)

with an R2 of 0.18. Volatility of GDP growth contributes to inflation volatility, as does a high central bank governor turnover rate while more open economies have more predictable inflation rates. Increased volatility of monetary growth rates and interest rates goes hand in hand with increased volatility of inflation. Controlling for these conditioning variables, pegged exchange rate regimes exert a negative but insignificant effect on inflation volatility. If the volatility of monetary growth is excluded from the regression, the coefficient on the regime dummy increases and becomes significant, which can be interpreted as evidence that pegged regimes exert a disciplinary effect not only on the mean but also on the volatility of monetary growth rates.

The negative coefficient on interest rate volatility presumably reflects a “transfer” effect: by ruling out adjustments in nominal exchange rates in response to shocks, pegged regimes transfer the pressure to other variables, including interest rates and reserves. The effect on the latter is examined in the last columns of Table 3, revealing a substantially larger volatility of reserves under pegged rates, a feature which remains robust to controlling for terms of trade and GDP volatility.

IV. Growth

We next turn to the comparative growth performance under alternative nominal exchange rate regimes. In contrast to the expansive literature on inflation, economic theory offers relatively few sharp predictions about the link between the nominal exchange rate regime and economic growth. To discuss the potential linkages, it is conceptually useful to first distinguish between variations in output (GDP) that arise from changes in the use of unemployed resources (that is, variations around the “full employment” level of output) and growth of full employment output itself.

In a world with other nominal rigidities, flexibility of the nominal exchange rate may be expected to facilitate adjustment and to restore full employment following adverse shocks. One might, thus, expect the variability of output and employment to be lower under flexible exchange rate regimes. The flexibility of prices, however, is itself likely to be a function of the exchange rate regime. Thus, it has been argued that the adoption of pegged exchange rates will reduce wage and price stickiness precisely because it sharply limits the ability of the government to counteract the adverse effects of stickiness in the presence of shocks.

Moving beyond business-cycle variations, pegged exchange rate regimes have been argued to foster faster economic growth [Aizenman (1991), Ghosh and Pesenti (1994)]. From simple growth accounting, any such effects must either influence the rate of factor accumulation, that is, investment and employment growth, or the growth in total factor productivity. Growth effects from increased rates of factor accumulation have been argued to arise mainly through higher investment under pegged rates, a result of a reduced real risk premium reflecting the higher policy credibility we identified above. An impact of the exchange rate regime on total factor productivity growth may arise either through an effect on the speed of sectoral adjustment to shocks or through a link between the regime and trade growth or economic openness in general, which in turn have long been argued to stimulate productivity growth through a variety of channels. Both channels, and in particular the link between exchange rate flexibility and the growth of trade remain, however, quite controversial. 20/

Accordingly, we turn again to empirics to throw some light on the importance of alternative channels. The remainder of this section is organized much as section 3, above. Subsection 1 reports basic results on growth performance across various regimes. Subsection 2 turns to regression analysis to examine whether differences in growth rates across regimes may be attributed to different investment rates, different productivity growth related to growth of international trade, or different residual productivity growth. Subsection 3 presents a simultaneous equation model, in order to check the robustness of the results when the regime is endogenous. Subsection 4 turns to the business cycle dimension and studies the volatility of growth and employment under alternative exchange rate regimes.

1. Unconditional means

Over our full sample of 3732 observations, GDP growth averaged 3.7 percent per year; almost 4 percent in the industrialized and upper-middle income countries and 3.6 percent in the lower-middle income and low-income countries. The fastest growth rates were experienced under the pegged exchange rate regimes, which averaged 4 percent growth per year, compared to 2.7 percent for the intermediate regimes, and 2.6 percent under flexible exchange rate regimes. Controlling for time factors by subtracting the annual means (see Table 4) reveals that most of these differences can be attributed to the higher global growth rates in the 1960s. Taking out the annual means reduces the difference between fixed and floating rates to less than 1/4 percent per year. The asymmetry of the pure float regime again emerges: while high-income countries under pure floats experienced (marginally) above-average growth rates, low-income economies with floating rates experienced the lowest growth rate of all regimes.

Table 4:

Average GDP Growth

(Deviation From Annual Global Mean)

article image

2. Conditional means

We next examine the partial effect of the nominal exchange rate regime on growth controlling for other growth determinants, including time dummies, the variability of the terms of trade σ(Δ(TT)) (calculated as a three year centered moving standard deviation of the terms of trade), lagged growth in government consumption as a readily available fiscal stimulus proxy (Δg), the investment to GDP ratio (i/y), the growth rate of trade (ΔTrade), measured as the growth rate of exports plus imports, and the World Bank’s index of development (y). The latter is coded in decreasing order of development, hence conditional convergence would be reflected by a positive coefficient. 21/

a. OLS results

To begin with, we only include those controls which are likely to be fairly exogenous to the exchange rate regime: the annual dummies, terms of trade variability, and the lagged government consumption variable. The results for the exchange rate regime dummy are given in Table 5 under the heading “Regression 5,” revealing a moderate positive association between economic growth and the flexibility of the exchange rate regime, with an average growth difference of 0.3 percent per annum between countries on flexible and fixed rates.

Table 5:

Growth Regressions

article image
article image