Exchange rate regime choice in Central and Eastern European transitional economies.

Klyuev, Vladimir

I. Introduction

The choice of exchange rate regimes by Central and Eastern European (CEE) transitional economies exhibits a surprising degree of heterogeneity both across countries and over time. The region features several currency boards, fixed but adjustable pegs, crawling pegs, bands, and crawling bands, as well as almost pure floats. A look at the evolution of the exchange rate regimes summarized in the Appendix reveals that this evolution does not seem to have a particular direction. The Czech Republic, Poland and a number of other countries fixed their exchange rates at the beginning of the transition, but later gradually moved from pegs to more flexible regimes. Romania has had a floating regime since 1991, but the extent to which the exchange rate was managed first decreased and then increased. The Baltic states fixed their exchange rates within a couple of years after obtaining independence. Bulgaria introduced a currency board after several years of floating.

This diversity seems surprising given the commonality of the communist legacy and the shared vision of the future where all countries in the region proclaim that they are building market economies and democratic political regimes with membership in the European Union as their goal. True, these countries are distinct in many aspects, but they hardly differ more from each other than Western European countries, which have maintained broadly similar arrangements for most of this century.

The traditional economic literature developed from Mundell's [1961] and McKinnon's [1963] seminal work on optimal currency areas does not help explain the diversity of the choices. All small and open economies, located in the same geographic region, similarly endowed with natural resources, the CEE countries should all have opted for the same regime. More sophisticated versions of the optimal currency area literature focus on the statistical distributions of various shocks (e.g., Tumovsky [1976]). The applicability of this approach to explaining the regime choice is limited since, by the very nature of transition, the countries of the region are not in steady-state positions, and the distributions of shocks cannot be considered stationary. Hence, the past offers little guidance with regard to future disturbances, so there is no relevant history that would allow one to apply this criterion.

The "new" theory of optimal currency areas, whose main advances are summarized by Tavlas [1993], offers more insights into the choices made by transitional economies. It allows for rational expectations and brings political economy into the picture. As one implication, it emphasizes the use of a fixed exchange rate as a nominal anchor in fighting inflation. This issue is very relevant for transitional economies, since price liberalization at the beginning of transition resulted in a powerful outburst of inflation and in some economies it started a wage-price inflationary spiral (Bruno, 1993). Moreover, since not all regulated prices were liberalized at the same time, increases in controlled prices provided further inflationary shocks. An introduction of indirect taxes had a similar effect. In addition, the rapid economic transformation of these countries required continual relative price adjustment, which was inflationary in the presence of downward price rigidity (Coorey et al, 1998). Finally, lax fiscal policies, in particular the support of loss-making state-owned enterprises, in some cases resulted in budget deficits financed by money creation (Fischer et al, 1996). (1) As a result, all the transitional economies confronted the issue of curbing runaway inflation, even though the extent of the problem was clearly different in different countries.

Fighting inflation requires a nominal anchor, and the choice is generally between the money supply and the exchange rate. Calvo and Vegh [1999] offer an excellent survey of the issues surrounding this choice. An important advantage of the exchange rate is that it provides a highly visible, easily verifiable target, while monitoring the behavior of the money supply is quite complicated. Such a visible anchor helps coordinate the expectations of price- and wage-setters around a low-inflation equilibrium. It also serves to reinforce the government's commitment to the stabilization effort. Monetary and fiscal policies inconsistent with maintaining the exchange rate target would result in a collapse of the regime on which the government has staked its credibility and consequently damage its electoral prospects. A fixed exchange rate regime facilitates the rebuilding of real money balances by economic actors, which will be demanded if the inflation rate is expected to go down. On the contrary, a reduction in the rate of growth of the money supply under a money-based stabilization will leave businesses and households starving for liquidity, which will drive interest rates up and plunge the economy into a recession. (2) Providing just enough cash at the beginning of the program (a one-time jump in the level of the money supply, followed by a reduction in its rate of growth) is problematic for two reasons. First of all, the credibility of a money-based stabilization program that starts with an expansion of domestic credit will be very much in question. In addition, the unpredictability of money demand in transitional economies makes the calculation of the right adjustment a highly precarious exercise, while the unstable money multiplier and underdeveloped indirect instruments of monetary policy make it difficult to meet a given target for broad monetary aggregates.

The superiority of the exchange rate anchor is by no means uncontested. Calvo and Vegh [1999] note that the choice between the two nominal anchors involves a tradeoff of "recession now" (money-based stabilization) versus "recession later" (exchange-rate-based stabilization); the authors discuss both empirical evidence to this effect and conceptual reasons why this may be so. Tornell and Velasco [1995, 1998] cast doubt on the assumption that fixed exchange rate regimes impose more fiscal discipline on the government. They note that fiscal laxity will undermine a peg only after some time, forcing a discrete devaluation in the future. Under a float, budget deficits financed by money creation will lead to an immediate depreciation. If the value of the domestic currency, and the overall price level, which is linked to it, affect private citizens' welfare and hence their level of support for the government, the latter will face the consequences of fiscal profligacy later under a fixed exchange rate regime and may well opt for higher spending under a peg than under a float, if its time horizon is sufficiently short.

In spite of these reservations, the belief in the efficacy of the exchange rate as a nominal anchor seems to have dominated the thinking of economic policymakers and external advisors at the outset of transition. Statements to this effect permeate such edited volumes as Williamson [1991], Barth and Wong [1994], and Sweeney et al. [1999], as well as numerous other books and articles. Floating was grudgingly recommended to countries that lacked sufficient foreign exchange reserves, where the alternative was seen to be infeasible.

On the other hand, it is well known that enlisting the support of an exchange-rate anchor to fight inflation is not without problems. Most notably, the inflation rate does not drop immediately to trading-partner levels, for a variety of reasons, including imperfect credibility and explicit or implicit backward wage indexation. The result is a real appreciation of the domestic currency, which undermines the competitiveness of the country's exports and encourages imports. This causes the trade balance and the current account balance to deteriorate and raises the question of sustainability of this type of stabilization in the long run. Indeed, the countries of the region that have chosen to peg have typically seen a deterioration in the trade and current account balances. It may be argued that restructuring and productivity growth should lead to an appreciation of the equilibrium real exchange rate, and that the current account deficits are comfortably financed with capital inflows. At the same time, the accounts of policy discussions in CEE countries indicate that real appreciation, loss of competitiveness, and current account deficits have been perceived as a problem, particularly in the wake of the Mexican and the Asian financial crises. Hence, having reduced inflation to moderate levels, a number of countries in the region (e.g., the Czech Republic, Hungary, and Poland) have introduced more flexible arrangements.

This article offers a simple model of exchange rate regime determination where the tradeoff between the use of the exchange rate as an anti-inflation tool and as a competitiveness tool takes center stage. While the discussions of the optimal choice of exchange rate regime abound, relatively few formal models of this choice have been developed in the literature. Our model is related to those where price stability (facilitated by a fixed exchange rate) is traded off against a stimulus to the real economy that may be produced by depreciation of domestic currency. (3) Those models (e.g., Devarajan and Rodrik [1992]; Edwards [1996]) approach the choice of exchange rate regime from the perspective of long-term optimality. The policymakers choose a regime that will minimize the expected value of their loss function, while the economy is subject to shocks with a known probability distribution. (4) There are no linkages between periods.

A key feature of our model is a link between past inflation and the present, which comes in the form of backward wage indexation. In addition, we take a shorter-run approach in view of the fluidity of the choice of exchange rate regimes and the shortness of policymakers' horizons in the economies in transition, especially in the earlier years. (5) The regime is chosen for one period only, when past inflation is known, so the choice is optimal ex post.

Our model predicts a nonlinear relationship between the inflation rate and the degree of exchange rate flexibility. The optimal degree of flexibility first increases with the rate of inflation, reflecting the concern for maintaining external competitiveness. At very high rates, however, inflation is seen as the most important problem, and the use of a fixed exchange rate as a nominal anchor is called for. A set of regressions confirms that this relationship indeed exists and is fairly robust in the data. In the recent literature a similar nonlinear relationship between inflation and exchange rate flexibility has been found for Latin America (Collins [1996]; Frieden et al. [2000]). While the justification for expecting high-inflation countries to opt for a fixed exchange rate regime is the same nominal anchor argument that we use, these papers do not convert the logic into a formal model.

Our model is an attempt to capture in a parsimonious form the essential features of the actual choice of exchange rate arrangements made by policymakers in the countries of Central and Eastern Europe. While the optimal regime is derived from a minimization of an explicit loss function, no direct link is made between this loss function and the welfare of the citizens of those countries. This is a positive description of the choices made rather than a normative model of what the optimal exchange rate regime should be.

While the tradeoff between inflation and real appreciation is relevant for the countries in the Commonwealth of Independent States (CIS) as well, we do not see our model as adequately describing their choice of exchange rate regime. The main reason is that the model assumes consistency between the exchange rate regime on the one hand and monetary and fiscal policies on the other, and such consistency is generally lacking in the CIS countries. Moreover, an important premise of our theory--that a fixed exchange rate is used as an instrument of choice in inflation stabilization programs--is not borne out by evidence in that region. In addition, the CIS governments have been unable to commit credibly to tight monetary and fiscal policies that would make a peg sustainable. Important reasons for the difference between the CIS and CEE countries may be the lack of a natural anchor currency for the former, given their geographic location, their trade patterns, and the inadequacy of foreign currency reserves.

The article is organized as follows. The next section introduces the model, derives the optimal choice of exchange rate regime, and discusses some comparative statics results. Then we study the dynamic implications of the model for the evolution of inflation and exchange rate regimes in the region. We discuss the limitations of the model in Section III. Empirical results along with a number of robustness checks are presented in Section IV.. The last section concludes.

II. The Model

The discussion in the introductory section singles out the nominal anchor property of a fixed (or, more generally, pre-announced) exchange rate, inflation inertia, and concern about real appreciation as issues relevant for the choice of exchange rate regime in transitional economies. A parsimonious model of an open economy that highlights these features has been developed by Edwards [1993], and we take it as a foundation for our political economy model.

The structure of the economy is described by equations (1)-(6) below.

(1) [[pi].sub.t] = [alpha][[pi].sub.Tt] + (1-[alpha])[[pi].sub.Nt]

(2) [[pi].sub.Tt] = [d.sub.t]

(3) [d.sub.t] = [phi][[pi].sub.t-1], 0 [less than or equal to] [phi] [less than or equal to] 1

(4) [N.sup.D] ([P.sub.N] / [P.sub.T], [Z.sub.t]) = [N.sup.S] (W / [P.sub.N])

(4') [eta] ([[pi].sub.Nt]- [[pi].sub.Tt])+ [delta] [z.sub.t] = [epsilon]([w.sub.t] - [[pi].sub.Nt]), [eta] < 0, [delta] > 0, [epsilon] < 0

(5) [w.sub.t] = [gamma][[pi].sub.t-1] + (1-[gamma])[[pi].sup.e.sub.t], 0 [less than or equal to] [gamma] [less than or equal to] 1

(6) [[pi].sup.e.sub.t] = [[pi].sub.t]

The economy produces two types of goods: a tradable and a nontradable commodity. Equation (1) defines overall inflation as a weighted sum of the rates of tradable and nontradable price increases. Equation (2) states that purchasing power parity holds for tradable goods. Hence, an increase in the price of tradables equals the rate of devaluation. Equation (3) defines the exchange rate regime of the country as a passive crawling peg. Parameter [phi] determines the extent to which devaluation compensates for past inflation, with [phi] = 0 corresponding to a fixed exchange rate regime and [phi] = 1 corresponding to real exchange rate targeting. Equation (4) states that the demand for nontradables, expressed as a function of their relative price and aggregate domestic demand [Z.sub.t], must equal the supply of nontradables, which depends upon the real product wage in that sector. Equation (4') is obtained by differentiating equation (4) with respect to time. Here [eta] is the demand elasticity of nontradables with respect to their relative price; [delta] is the demand elasticity of nontradables with respect to aggregate demand pressures; and e is the supply elasticity of nontradables with respect to the real product wage. Equation (5) says that the growth rate of the nominal wage is a weighted average of past and expected future inflation, and the parameter [gamma] captures the degree to which wage formation is backward-looking. Finally, equation (6) reflects the assumption that inflationary expectations are formed rationally.

It is important to note that partial backward indexation of wages should be seen as a means of introducing inertial elements into the model rather than a way to capture a particular institutional mechanism. Wage indexation may be implicit rather than explicit. Moreover, inflation may have inertia because the exchange rate anchor is not completely credible, or because inflationary expectations persist (some sort of adaptive expectations). Any of these factors would make the rate of inflation higher than the rate of depreciation and result in real appreciation.

Equations (1) through (6) can be manipulated to obtain the following autoregressive process for inflation:

(7) [[pi].sub.t] = a[[pi].sub.t-1] + b[z.sub.t],

where a = ([eta]+[epsilon][alpha])[phi]+[epsilon](1-[alpha])[gamma] / ([eta]+[epsilon][alpha])+[epsilon](1-[alpha])[gamma] and b= -[delta](l-[alpha]) / ([eta]+[epsilon][alpha])+[epsilon](1-[alpha])[gamma],

0 [less than or equal to] a [less than or equal to] 1, b>0.

Expansionary policies (positive [z.sub.t],) fuel inflation, while a contraction helps break the inflationary cycle. In this article we want to focus on nominal anchor properties of the exchange rate, so we abstract from aggregate demand management, or rather, posit that monetary and fiscal policies are strictly in line with the chosen exchange rate path. We discuss this important assumption in detail in Section IV. Here we set [z.sub.t] = 0, and (7) becomes

(7') [[pi].sub.t] = a[[pi].sub.t-1].

We will also have:

(8) [[pi].sub.Tt] = [d.sub.t] = [phi][[pi].sub.t-1], and

(9) [w.sub.t], = [gamma] [[pi].sub.t-1] + (1 - [gamma])a[[pi].sub.t-1],

so the rate of real appreciation is

(10) [s.sub.t], = [w.sub.t] - [[pi].sub.Tt] = [gamma](1 - [phi]) x [eta]+[epsilon] / ([eta] + [epsilon][alpha])+[epsilon](1-[alpha])[gamma] [[pi].sub.t-1].

Note that real appreciation is identified with an increase in the product wage in the tradable sector and reflects a loss of competitiveness in the international market. (6)

If [delta] = 1, then a = 1 and [s.sub.t]=0. The inflation process has a unit root (there is no nominal anchor), and there is no real appreciation. Of course, in reality a break on inflation may be provided by policies other than a fixed exchange rate. These policies would be reflected in the [z.sub.t] term.

Having [delta] = 0 (credibly fixed exchange rate) is not enough to halt inflation in its tracks unless [gamma] = 0. If the wage formation process is backward-looking, inflation will not drop to zero immediately, and there will be some real appreciation.

Equations (7') and (10) describe the dynamics of inflation and real appreciation for given parameter values and the initial inflation rate. In keeping with our focus on the choice of exchange rate regime, we now make [delta] (the degree to which the exchange rate accommodates past inflation) a decision variable. We assume that the authorities minimize the following loss function:

(11) L = [[pi].sup.2.sub.t] + k[([s.sub.t] - [s.sup.*].sup.2] + m([[phi].sub.t]- [[phi].sub.t-1], k>0, m>0.

This reflects an aversion to inflation, an unwillingness to have a real appreciation (and hence lose competitiveness in the tradables sector), and the cost of changing exchange rate arrangements. As is typical in the political economy literature, the loss function is assumed to be a convex function of its arguments, so that, e.g., a given increase of inflation is perceived to be more onerous if it starts from a higher base. The preferred level of inflation is zero, while the preferred rate of real appreciation is negative ([s.sup.*]<0). (7) This assumption can be seen as a shortcut combining two common suppositions (e.g., Devarajan and Rodrik [1992]); namely, that real depreciation has an expansionary effect and that the government's real output target is greater than the natural rate. Changing the regime is presumed to be costly because of the status quo bias that has been well documented in the political economy literature in various incarnations (e.g., Alesina and Drazen [1991], Fernandez and Rodrik [1991]). Changing an important rule by which the economy functions (and even an anticipation of such a change) introduces uncertainty and disruption in economic life, so this decision is not taken lightly. Moreover, in view of the argument that the choice of exchange rate regime has distributional consequences, an attempt to change the regime would generate resistance on the part of negatively affected groups; therefore, it is unlikely to be undertaken even by a benevolent government unless the net welfare benefit of such a change is high (Rodrik [1994]). These institutional costs are incurred regardless of the direction of the change, which explains the square term. The assumption that the cost of the exchange rate regime adjustment depends upon the magnitude of adjustment reflects the fact that marginal changes of the regime (changing the rate of the crawl or playing with the width of the band) are likely to introduce less disruption and generate less resistance than a radical switch in the regime.

The authorities choose the regime [[phi].sub.t], to minimize L taking [[phi].sub.t-1], and [[pi].sub.t-1] as given. L is a quadratic function of [[phi].sub.t] with a positive coefficient on the square term, so it is convex in [[phi].sub.t] and the point where the derivative of L with respect to [[phi].sub.t], equals zero is the global minimum. Of course, we should also take into account that [[phi].sub.t] is only allowed to vary between zero and one. The first order condition yields the following expression:

(12) [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII]

This is a nonlinear function of [[pi].sub.t-1]. If [[pi].sub.t-1] = 0, then [[phi].sub.t] = [[phi].sub.t-1], and there is no reason to change the regime in this framework. It is obvious that when [[pi].sub.t] is small, [[phi].sub.t] is an increasing function of [[pi].sub.t-1]. For large [[pi].sub.t-1], [phi], tends to a limit that is less than one and could be negative. Of course, [[phi].sub.t] < 0 is not feasible; thus, for any inflation rate above a certain limit, fixing the exchange rate may be the optimal response. Further analysis shows that the expression in (12) first increases and then decreases with [[pi].sub.t-1], so the model validates the logic of our informal argument. (8)

Holding everything else constant, greater weight in the loss function on real appreciation (increasing k) favors a more accommodative stance (higher [[phi].sub.t]). More weight on inflation (reducing both k and m) favors smaller [[phi].sub.t], and increasing the cost of adjustment, m, makes the regime more sticky.

A greater degree of backward indexation of wages, [gamma], increases [kappa], [zeta], and [theta], and decreases [xi]. The effect on the exchange-rate regime is ambiguous and depends on parameter values and initial conditions. There is no effect at zero inflation. At small rates, a higher degree of indexation will be accommodated by choosing a more flexible regime. At high rates of inflation, depending in particular on the weight assigned to external competitiveness in the loss function and on the regime in place at the moment, greater backward indexation may be offset with more aggressive pegging, may be accommodated through more flexibility, or may have no effect on the choice of regime if the choice is already at one of the extremes.

For [gamma] < 1, [kappa] is a decreasing function of [alpha]. Hence, for low inflation (such that the quadratic terms in (12') can be ignored), less flexibility in the exchange rate is preferred by more open economies (since [alpha] is the share of tradables in the consumer basket) for any given [[pi].sub.t-1]. [theta] declines with [alpha]. How that affects the choice of regime depends on parameter values and initial conditions. (9) [xi] is an increasing function of [alpha], so the term that contains x contributes to making the regime less flexible for more open economies. [xi] is a complicated function of [alpha]. One can show (see Klyuev [2000]) that [differential][zeta] / [differential][alpha]<0 whenever |[eta] + [epsilon][alpha]| > [epsilon][gamma](1-[alpha])|. The last inequality will hold except in quite closed economies with a high degree of wage indexation (low [alpha], high [gamma]). Therefore, the term [zeta][[pi].sup.2.sub.t-1] in the numerator somewhat mitigates the tendency of more open economies to have more sticky exchange rates, but it acquires significance only when the inflation rate is quite high. All in all, one can expect trade openness to be associated with less exchange rate flexibility at low rates of inflation, while at higher rates the result may be reversed.

III. Dynamic Implications

In the model developed above the policymakers look only one period ahead--an assumption that could be justified by uncertainty about more remote horizons, be it uncertainty about economic developments, geopolitical developments, or simply about the chances of being reelected. It might be interesting, however, to see what kind of dynamics the model implies for inflation and exchange rate regimes over several periods if the system suffers no major disturbances. Specifically, the following experiment is performed. We fix the structure of the economy as described by equations (1) through (6), choose a starting point (initial exchange rate regime and initial inflation), and let a sequence of governments, which minimize one-period loss function (11), determine the time path of the economy. We assume there are no shocks along the way. Each government will choose the regime [??] as in (12), given the regime and the inflation rate it has inherited from the previous government, and will leave this regime and inflation determined by (7') as a legacy to its successor.

Equations (7') and (12) determine the time path of inflation and the exchange- rate rule from any point onward, provided the loss function does not change and the system is not subject to any shocks. A phase diagram can be used to analyze the behavior of the system.

Inflation is stationary in two cases: when inflation is zero or when the exchange rate regime parameter [??] equals one. If [??], is less than one, inflation decreases in period t.

Equation (12) can be manipulated to obtain a change in the degree of exchange rate flexibility between two periods:

[[??].sub.t] - [[??].sub.t-1] = [kappa] + ([theta] - [zeta])[[pi].sub.t-1] - ([theta] + [xi]) [p.sup.2] [[pi].sub.t-1] [[??].sub.t-1] / 1 + ([theta] + [xi])[[pi].sup.2.sub.t-1] / [[pi].sub.t-1]

The degree of flexibility does not change in two cases: when inflation is zero or along the downward-sloping line

[??] = [kappa] / ([theta] + [xi])[pi] + ([theta] - [zeta]) / ([theta] + [xi])

Above this line, flexibility decreases over time. Below, flexibility increases provided inflation is still above zero. The phase diagram is shown in Figure I. If the parameter [??] is forced to lie between zero and one, the system will end up in one of a continuum of steady states. The economy may converge to zero inflation, and then it will keep the regime at which inflation first hit zero. Alternatively, it may converge to the fully flexible regime ([??] = 1) and an inflation rate between zero and [kappa]//([xi]+[zeta]. (10)

(FIGURE I OMITTED]

Figure II displays the results of simulating the dynamic path of a fictional economy described by our model. The starting point was chosen to represent a typical situation of a Central European economy at the beginning of transition with high inflation and a flexible exchange rate regime. The two panels present cases corresponding to two possible steady states: a flexible exchange rate regime at low inflation (Panel a) and a zero inflation at an intermediate regime (Panel b). Irrespective of the eventual steady state, a notable feature of transitional dynamics is a dramatic initial decrease in the degree of exchange rate flexibility, followed by a gradual increase in flexibility, while inflation decreases monotonically. This kind of dynamics fits very well the stylized description of the evolution of exchange rate regimes and inflation in a large number of Central and Eastern European countries (Koch [1997], Masson [1999]), as well as the path suggested by prominent western advisors (e.g., Sachs [1996]).

[FIGURE II OMITTED]

IV. Discussion and Limitations of the Model

Certainly a simple model cannot capture all the complexity surrounding the issue of exchange rate policy. In this section we will try to address the most obvious questions that arise vis-a-vis the model.

Credibility

The model does not distinguish between credible and incredible policy announcements. The exchange rate rule is assumed to be known to the public and followed by the government. In reality, public pronouncements are not always reliable, and rules are not always obeyed.

While recognizing the importance of the issue of credibility, we circumvented it in our model so that we could focus in a tractable way on the central point of the argument: the conflict between the internal and external balance, or the trade-of-between inflation and real appreciation. To a degree, imperfect credibility of exchange-rate announcements is captured in backward indexation of wage contracts.

Choice of policy rule

An important question is how one can quantify exchange-rate policy to move from description to a tractable model, while still reflecting the multitude of possible arrangements. The prevalent answer is to reduce the diversity to just two regimes. Most commonly, the countries are assumed to face a choice between a flexible regime and a fixed regime with (Edwards, 1996) or without (Devereux and Engel, 1998) the possibility of devaluation or abandoning the peg. Occasionally, some version of a managed float is pitted against a peg, immutable or not (Collins, 1996). This dichotomy serves poorly our objective of accounting for the tremendous diversity of exchange rate arrangements in Central and Eastern Europe and tracing the evolution of exchange rate regimes within individual countries, which we read as a history of mostly marginal adjustments with occasional sharp breaks. Conceptually, we much prefer a continuous measure of exchange rate regime, even though for an empirical implementation we will have to go to a cruder scale. (11)

The question still remains, of course, whether this particular continuous variable is an adequate representation of exchange-rate regime. The policy variable f was defined as the fraction of the previous period's inflation that the monetary authority was willing to accommodate through devaluation. Certainly, no country formulates its foreign exchange policy by announcing its [??]. The actual choices may include using foreign currency as legal tender; introducing a currency board; announcing a peg to a foreign currency or a basket of currencies in a less rigid way; announcing a crawling peg with a particular rate of crawl; specifying a fluctuation band; combining the latter two arrangements in a crawling band; a managed float without specifying (or committing to) an exchange rate target; or an independent float. One would hardly dispute associating [??] = 0 with a peg or a currency board. A value of [??] between zero and one can be thought of as describing a crawling peg. Of course, in the actual crawling pegs, the rate of crawl is specified directly rather than as a fraction of past inflation, but the correspondence is fairly close. One just has to be careful not to interpret a reduction in the rate of crawl, as inflation declines, as a change in regime. What crawling pegs attempt to achieve is to provide a nominal anchor (pre-announced path of the exchange rate) to focus inflationary expectations while accommodating inflation inertia. What differentiates crawling pegs is how aggressive they are in their anti-inflationary stance, which means how small the rate of crawl is relative to inflation registered in the past.

The correspondence between the model and reality is much looser when one brings wide currency bands or floating regimes into the picture. In the model, the path of the exchange rate is always pre-announced. In the discussion, we have associated higher values of [??] with a greater degree of flexibility. The logic for this assignment is that in the real world (at least in transitional economies during the initial stabilization), a lower degree of exchange rate flexibility is typically associated with a tougher stance on inflation, at the risk of allowing real appreciation. In the model, this stance is conveyed by a lower [??].

Aggregate demand management

In the original Edwards [1993] model, inflation could be brought down without using the exchange rate as a nominal anchor. Instead, the authorities could suppress aggregate demand, which would be reflected in a negative [z.sub.t] term in equation (7). Our neglect of this term amounts to assuming that countries that pre-announce the path of the exchange rate conduct policies consistent with maintaining this path. In addition, given our interpretation of [??] as a degree of exchange rate flexibility, we are presuming that countries that opt for a flexible regime do not try to bring inflation down through other means.

Certainly neither of the assumptions is true in every case, but individual deviations would be captured in the error term, and we do not expect the deviations to have a systematic pattern. First, for reasons noted in the introduction, money-based stabilization programs have rarely been attempted in Central and Eastern Europe. Slovenia may be a major exception, but even there the exchange rate was not allowed to float freely. Secondly, it has been noted that among the transitional economies of Central and Eastern Europe, those with more sound fiscal policies tended to adopt a pegged exchange rate (Begg, 1998). Whether the reason for that correlation is that less disciplined countries realized that exchange rate pegs would not be sustainable and did not even try to launch such an infeasible regime, or whether the story of a fixed exchange rate tying government's hands has some validity, this fact allows us to use the exchange rate regime as a sufficient statistic for government policy and reduces the decision making of the government to a uni-dimensional problem. On the other hand, it is extremely important to ascertain which story is the right one before attempting to formulate the government's problem, since if the decision concerning the choice of exchange rate regime is driven by the state of public finances, rather than the other way around, this would impose a constraint on the government that is not recognized in our model. Romania may be a case in point. It is a high-inflation country that should have embarked on an exchange-rate based stabilization program according to the model, but it has continually been unable to muster the necessary macroeconomic discipline, which in particular has been reflected in a sorrowful state of public finances. One may easily question institutional capacity in other Southern European countries. Still, on the whole, in Central and Eastern Europe macroeconomic policies appear to be relatively consistent and governance is adequate compared to countries further east, where the application of the model would seem to be unwarranted.

Currency boards

One would be right to argue that a currency board is a type of exchange-rate regime quite different from a simple peg. The principal difference is the degree of institutional commitment to maintaining the peg and the consequent difficulty of changing the regime, which is much greater under a currency board. This translates into an empirical fact that currency boards are much more stable than other regimes. In particular, no country in Central and Eastern Europe has ever revoked this arrangement. (12) The model would handle the introduction of a currency board, which follows an episode of high inflation, as decreasing the rate of crawl, [??], to zero and a rise in the cost of changing the exchange rate regime, m. A sharp increase in m is a natural way in the model to capture what is special about a currency board--the institutional difficulty of changing this arrangement, which will generate, albeit somewhat tautologically, the persistence of such a regime (both in real life and in the model). What the model does not explain is why some countries choose ordinary pegs in the face of high inflation while others go all the way to currency boards.

V. Empirical Analysis

The purpose of this section is to demonstrate the validity of the model by comparing its predictions with the actual behavior of exchange rate regimes. We use panel data for the years 1990-98 for 13 Central and Eastern European transitional economies: Albania, Bulgaria, Croatia, the Czech Republic, Estonia, Hungary, Latvia, Lithuania, Macedonia, Poland, Romania, Slovakia, and Slovenia. This exhausts the list of countries in the region for which adequate data are available. The year 1990 is taken to be the starting year of the economic transformation in the region. Before that, the developments were mostly on the political front and centered on demonopolizing political power and gaining de facto independence from the Soviet Union. The foreign exchange regimes were largely unreformed, and currencies were not convertible. Of course, not all of the countries existed in 1990, so for them the series start with independence.

The most serious conceptual difficulty to overcome when testing the model against the data is finding an appropriate proxy for the dependent variable. The first conceptual peril, as discussed above, is to associate the policy parameter, f, in the model with the degree of exchange rate flexibility. Once this leap of faith is made, the remaining steps are classifying the existing regimes into a number of categories and arranging the categories in the order of increasing flexibility.

We mostly rely on the IMF's publications to classify exchange rate regimes. The monthly issues of International Financial Statistics feature a table of exchange rate arrangements. Until 1999, this classification contained three categories relevant for Central and Eastern European economies: pegged to a currency or a basket of currencies; managed float; independent float. We use the IMF classification as of December 31 of a given year and assign the value of 0, 1, or 2, respectively, to the dependent variable for the three regimes. The variable increases with the degree of flexibility.

This classification is quite coarse, and we have constructed an alternative coding for the dependent variable based on our reading of the IMF's reports on Exchange Arrangements and Exchange Restrictions, as well as on the Economist Intelligence Unit's Country Profiles and Country Reports. The coding, in the order of increasing flexibility, is: 0- currency board; 1--peg; 2--crawling peg; 3--band; 4--crawling band; 5--managed float; 6--independent float. This classification is essentially the format currently used in the International Financial Statistics. While this classification scheme is much richer than the first one, its disadvantage is that the coding has been done by the author and hence may reflect some subjective bias. In any case, the two variables are highly correlated, and the empirical results turn out to be quite similar.

The main explanatory variables are the lagged value of the dependent variables, the rate of inflation, and inflation squared. The model suggests a nonlinear relationship between these and the dependent variable. Estimating a nonlinear equation presents high demands on the data and is hardly justified since the exact functional form depends on the specification of the loss function and is unlikely to be robust. The main message that we take from the model is that the desired degree of exchange rate flexibility first increases and then decreases with inflation. The exchange rate regimes are "sticky" and are not readily changed. The spirit of the model will be retained by simply putting a linear combination of the three explanatory variables on the right hand side.

Of course, the parameters of the economy and the weights in the loss function may change from country to country. Smaller foreign reserves would make a country more wary about real appreciation, so k in the loss function increases, or s* becomes more negative, or both, and the country will prefer a more flexible regime for any given rate of inflation. Similarly, higher unemployment may increase the target for output growth and make a country more averse to real appreciation, which would translate into a more flexible arrangement. More engagement in international trade corresponds to a greater a in the model. As the analysis indicates, at low levels of inflation a country heavily involved in trade will favor a fixed exchange rate regime, but it may prefer more flexibility when the inflation rate is high and when a fixed exchange rate would lead to a loss of competitiveness. To test this hypothesis, total trade (exports plus imports) as a share of GDP is included in the regression as a separate variable and interacted with inflation.

The authorities of countries saddled with large external debt tend to be wary of devaluation of their currency, as it would increase the debt burden expressed in domestic currency. On the other hand, servicing the debt requires earning foreign exchange, which is easier when the economy is more competitive. Given this trade-off, it is difficult to tell exactly how high external debt ratios would influence the "preferred" level of the exchange rate. What seems to happen in practice is that the first argument prevails for some time, as policymakers cling to the old exchange rate for fear of creating balance sheet problems, but eventually balance-of-payments difficulties arise, and the currency is allowed to depreciate. It should also be noticed that large step devaluations may occur without a change in the exchange rate regime (e.g., under an adjustable peg). Alternatively, devaluations are frequently associated with changes in the exchange regime, but not with a particular direction of the change. Indeed, countries commonly fix the exchange rate at a level more depreciated than the pre-existing floating rate, while a transition from a peg to a float often occurs under speculative pressure, which leads to depreciation as soon as the currency is floated. Because of this, we cannot predict the direction in which external indebtedness would influence the exchange-rate regime, but we include the ratio of external debt to GDP in the analysis to investigate the matter empirically.

Tables I and II give summary statistics for the sample for the baseline regression. The data for all non-categorical variables come from the International Financial Statistics of the IMF. As we can see, all three exchange rate regimes distinguished by the IMF are almost equally represented in the sample, with independent floating occurring somewhat less frequently than the two other options. The mean inflation rate in the sample is 68.1 percent per year, which is quite high. At the same time, the median inflation in the sample is much lower, 22.5 percent. The mean is drawn to the right by relatively rare occurrences of very high inflation. The countries in the region trade a great deal with the outside world. There is considerable variation within the sample in the level of international reserves, which ranges from very low to quite high magnitudes. So does external debt. The unemployment rate also varies substantially.

The ordered logit regression results are summarized in Table III. Our baseline regression (Column 1) includes inflation, inflation squared, and dummies for lagged regime. There is a considerable degree of stickiness in exchange rate regime choice, as indicated by a large and highly statistically significant coefficient on the lagged dependent variable.

The idea of a nonlinear relationship between the inflation rate and the degree of exchange rate flexibility is confirmed by the regression. Both coefficients have the predicted signs and are highly statistically significant. The pseudo R-squared of 56 percent is fairly decent for such a parsimonious specification. (13)

The partial relationship between inflation and the degree of exchange rate flexibility, as given by the baseline regression, peaks at the inflation rate of 300 percent per year. When inflation is lower, the relationship is positive: the probability of choosing a fixed regime diminishes, and the probability of choosing an independent float rises if the rate of inflation increases, ceteris paribus. In this range, external competitiveness is the dominant concern. Above 300 percent per year, more inflation is more likely to be followed by a tightening of the exchange regime. Internal stability comes to center stage, and the exchange rate is used as a nominal anchor.

Table IV shows how the regime would be chosen if inflation and the previous regime were its only determinants. At low inflation the regime is likely to remain unchanged, (14) while at very high inflation a peg becomes the preferred alternative. While these outcomes comply with the model, the cutoff points are higher than what we had expected. At the same time, this table traces quite well the experiences of high inflation countries, as demonstrated by Table V. (15) The table accurately predicts Romania's drift from a peg to a managed float and to an independent float, Lithuania staying with its independent float, Poland staying with a peg, and Bulgaria switching to a fixed regime developments, or simply about the chances of being reelected. It might be interesting, however, to see what kind of dynamics the model implies for inflation and exchange rate regimes over several periods if the system suffers no major disturbances. Specifically, the following experiment is performed. We fix the structure of the economy as described by equations (1) through (6), choose a starting point (initial exchange rate regime and initial inflation), and let a sequence of governments, which minimize one-period loss function (11), determine the time path of the economy. We assume there are no shocks along the way. Each government will choose the regime [phi] as in (12), given the regime and the inflation rate it has inherited from the previous government, and will leave this regime and inflation determined by (7') as a legacy to its successor.

Equations (7') and (12) determine the time path of inflation and the exchange- rate rule from any point onward, provided the loss function does not change and the system is not subject to any shocks. A phase diagram can be used to analyze the behavior of the system.

Inflation is stationary in two cases: when inflation is zero or when the exchange rate regime parameter [phi] equals one. If [phi], is less than one, inflation decreases in period t.

Equation (12) can be manipulated to obtain a change in the degree of exchange rate flexibility between two periods:

[[phi].sub.t] - [[phi].sub.t-1] = [kappa] + ([theta] - [zeta])[[pi].sub.t-1] - ([theta] + [xi]) [p.sup.2] [[pi].sub.t-1] [[phi].sub.t-1] / 1 + ([theta] + [xi])[[pi].sup.2.sub.t-1] / [[pi].sub.t-1]

The degree of flexibility does not change in two cases: when inflation is zero or along the downward-sloping line

[phi] = [kappa] / ([theta] + [xi])[pi] + ([theta] - [zeta]) / ([theta] + [xi])

Above this line, flexibility decreases over time. Below, flexibility increases provided inflation is still above zero. The phase diagram is shown in Figure I. If the parameter [phi] is forced to lie between zero and one, the system will end up in one of a continuum of steady states. The economy may converge to zero inflation, and then it will keep the regime at which inflation first hit zero. Alternatively, it may converge to the fully flexible regime ([phi] = 1) and an inflation rate between zero and [kappa]//([xi]+[zeta]. (10)

[FIGURE I OMITTED]

Figure II displays the results of simulating the dynamic path of a fictional economy described by our model. The starting point was chosen to represent a typical situation of a Central European economy at the beginning of transition with high inflation and a flexible exchange rate regime. The two panels present cases corresponding to two possible steady states: a flexible exchange rate regime at low inflation (Panel a) and a zero inflation at an intermediate regime (Panel b). Irrespective of the eventual steady state, a notable feature of transitional dynamics is a dramatic initial decrease in the degree of exchange rate flexibility, followed by a gradual increase in flexibility, while inflation decreases monotonically. This kind of dynamics fits very well the stylized description of the evolution of exchange rate regimes and inflation in a large number of Central and Eastern European countries (Koch [1997], Masson [1999]), as well as the path suggested by prominent western advisors (e.g., Sachs [1996]).

[FIGURE II OMITTED]

IV. Discussion and Limitations of the Model

Certainly a simple model cannot capture all the complexity surrounding the issue of exchange rate policy. In this section we will try to address the most obvious questions that arise vis-a-vis the model.

Credibility

The model does not distinguish between credible and incredible policy announcements. The exchange rate rule is assumed to be known to the public and followed by the government. In reality, public pronouncements are not always reliable, and rules are not always obeyed.

While recognizing the importance of the issue of credibility, we circumvented it in our model so that we could focus in a tractable way on the central point of the argument: the conflict between the internal and external balance, or the trade-of-between inflation and real appreciation. To a degree, imperfect credibility of exchange-rate announcements is captured in backward indexation of wage contracts.

Choice of policy rule

An important question is how one can quantify exchange-rate policy to move from description to a tractable model, while still reflecting the multitude of possible arrangements. The prevalent answer is to reduce the diversity to just two regimes. Most commonly, the countries are assumed to face a choice between a flexible regime and a fixed regime with (Edwards, 1996) or without (Devereux and Engel, 1998) the possibility of devaluation or abandoning the peg. Occasionally, some version of a managed float is pitted against a peg, immutable or not (Collins, 1996). This dichotomy serves poorly our objective of accounting for the tremendous diversity of exchange rate arrangements in Central and Eastern Europe and tracing the evolution of exchange rate regimes within individual countries, which we read as a history of mostly marginal adjustments with occasional sharp breaks. Conceptually, we much prefer a continuous measure of exchange rate regime, even though for an empirical implementation we will have to go to a cruder scale. (11)

The question still remains, of course, whether this particular continuous variable is an adequate representation of exchange-rate regime. The policy variable f was defined as the fraction of the previous period's inflation that the monetary authority was willing to accommodate through devaluation. Certainly, no country formulates its foreign exchange policy by announcing its [phi]. The actual choices may include using foreign currency as legal tender; introducing a currency board; announcing a peg to a foreign currency or a basket of currencies in a less rigid way; announcing a crawling peg with a particular rate of crawl; specifying a fluctuation band; combining the latter two arrangements in a crawling band; a managed float without specifying (or committing to) an exchange rate target; or an independent float. One would hardly dispute associating [phi] = 0 with a peg or a currency board. A value of [phi] between zero and one can be thought of as describing a crawling peg. Of course, in the actual crawling pegs, the rate of crawl is specified directly rather than as a fraction of past inflation, but the correspondence is fairly close. One just has to be careful not to interpret a reduction in the rate of crawl, as inflation declines, as a change in regime. What crawling pegs attempt to achieve is to provide a nominal anchor (pre-announced path of the exchange rate) to focus inflationary expectations while accommodating inflation inertia. What differentiates crawling pegs is how aggressive they are in their anti-inflationary stance, which means how small the rate of crawl is relative to inflation registered in the past.

The correspondence between the model and reality is much looser when one brings wide currency bands or floating regimes into the picture. In the model, the path of the exchange rate is always pre-announced. In the discussion, we have associated higher values of [phi] with a greater degree of flexibility. The logic for this assignment is that in the real world (at least in transitional economies during the initial stabilization), a lower degree of exchange rate flexibility is typically associated with a tougher stance on inflation, at the risk of allowing real appreciation. In the model, this stance is conveyed by a lower [phi].

Aggregate demand management

In the original Edwards [1993] model, inflation could be brought down without using the exchange rate as a nominal anchor. Instead, the authorities could suppress aggregate demand, which would be reflected in a negative [z.sub.t] term in equation (7). Our neglect of this term amounts to assuming that countries that pre-announce the path of the exchange rate conduct policies consistent with maintaining this path. In addition, given our interpretation of [phi] as a degree of exchange rate flexibility, we are presuming that countries that opt for a flexible regime do not try to bring inflation down through other means.

Certainly neither of the assumptions is true in every case, but individual deviations would be captured in the error term, and we do not expect the deviations to have a systematic pattern. First, for reasons noted in the introduction, money-based stabilization programs have rarely been attempted in Central and Eastern Europe. Slovenia may be a major exception, but even there the exchange rate was not allowed to float freely. Secondly, it has been noted that among the transitional economies of Central and Eastern Europe, those with more sound fiscal policies tended to adopt a pegged exchange rate (Begg, 1998). Whether the reason for that correlation is that less disciplined countries realized that exchange rate pegs would not be sustainable and did not even try to launch such an infeasible regime, or whether the story of a fixed exchange rate tying government's hands has some validity, this fact allows us to use the exchange rate regime as a sufficient statistic for government policy and reduces the decision making of the government to a uni-dimensional problem. On the other hand, it is extremely important to ascertain which story is the right one before attempting to formulate the government's problem, since if the decision concerning the choice of exchange rate regime is driven by the state of public finances, rather than the other way around, this would impose a constraint on the government that is not recognized in our model. Romania may be a case in point. It is a high-inflation country that should have embarked on an exchange-rate based stabilization program according to the model, but it has continually been unable to muster the necessary macroeconomic discipline, which in particular has been reflected in a sorrowful state of public finances. One may easily question institutional capacity in other Southern European countries. Still, on the whole, in Central and Eastern Europe macroeconomic policies appear to be relatively consistent and governance is adequate compared to countries further east, where the application of the model would seem to be unwarranted.

Currency boards

One would be right to argue that a currency board is a type of exchange-rate regime quite different from a simple peg. The principal difference is the degree of institutional commitment to maintaining the peg and the consequent difficulty of changing the regime, which is much greater under a currency board. This translates into an empirical fact that currency boards are much more stable than other regimes. In particular, no country in Central and Eastern Europe has ever revoked this arrangement. (12) The model would handle the introduction of a currency board, which follows an episode of high inflation, as decreasing the rate of crawl, [phi], to zero and a rise in the cost of changing the exchange rate regime, m. A sharp increase in m is a natural way in the model to capture what is special about a currency board--the institutional difficulty of changing this arrangement, which will generate, albeit somewhat tautologically, the persistence of such a regime (both in real life and in the model). What the model does not explain is why some countries choose ordinary pegs in the face of high inflation while others go all the way to currency boards.

V. Empirical Analysis

The purpose of this section is to demonstrate the validity of the model by comparing its predictions with the actual behavior of exchange rate regimes. We use panel data for the years 1990-98 for 13 Central and Eastern European transitional economies: Albania, Bulgaria, Croatia, the Czech Republic, Estonia, Hungary, Latvia, Lithuania, Macedonia, Poland, Romania, Slovakia, and Slovenia. This exhausts the list of countries in the region for which adequate data are available. The year 1990 is taken to be the starting year of the economic transformation in the region. Before that, the developments were mostly on the political front and centered on demonopolizing political power and gaining de facto independence from the Soviet Union. The foreign exchange regimes were largely unreformed, and currencies were not convertible. Of course, not all of the countries existed in 1990, so for them the series start with independence.

The most serious conceptual difficulty to overcome when testing the model against the data is finding an appropriate proxy for the dependent variable. The first conceptual peril, as discussed above, is to associate the policy parameter, f, in the model with the degree of exchange rate flexibility. Once this leap of faith is made, the remaining steps are classifying the existing regimes into a number of categories and arranging the categories in the order of increasing flexibility.

We mostly rely on the IMF's publications to classify exchange rate regimes. The monthly issues of International Financial Statistics feature a table of exchange rate arrangements. Until 1999, this classification contained three categories relevant for Central and Eastern European economies: pegged to a currency or a basket of currencies; managed float; independent float. We use the IMF classification as of December 31 of a given year and assign the value of 0, 1, or 2, respectively, to the dependent variable for the three regimes. The variable increases with the degree of flexibility.

This classification is quite coarse, and we have constructed an alternative coding for the dependent variable based on our reading of the IMF's reports on Exchange Arrangements and Exchange Restrictions, as well as on the Economist Intelligence Unit's Country Profiles and Country Reports. The coding, in the order of increasing flexibility, is: 0- currency board; 1--peg; 2--crawling peg; 3--band; 4--crawling band; 5--managed float; 6--independent float. This classification is essentially the format currently used in the International Financial Statistics. While this classification scheme is much richer than the first one, its disadvantage is that the coding has been done by the author and hence may reflect some subjective bias. In any case, the two variables are highly correlated, and the empirical results turn out to be quite similar.

The main explanatory variables are the lagged value of the dependent variables, the rate of inflation, and inflation squared. The model suggests a nonlinear relationship between these and the dependent variable. Estimating a nonlinear equation presents high demands on the data and is hardly justified since the exact functional form depends on the specification of the loss function and is unlikely to be robust. The main message that we take from the model is that the desired degree of exchange rate flexibility first increases and then decreases with inflation. The exchange rate regimes are "sticky" and are not readily changed. The spirit of the model will be retained by simply putting a linear combination of the three explanatory variables on the right hand side.

Of course, the parameters of the economy and the weights in the loss function may change from country to country. Smaller foreign reserves would make a country more wary about real appreciation, so k in the loss function increases, or s* becomes more negative, or both, and the country will prefer a more flexible regime for any given rate of inflation. Similarly, higher unemployment may increase the target for output growth and make a country more averse to real appreciation, which would translate into a more flexible arrangement. More engagement in international trade corresponds to a greater a in the model. As the analysis indicates, at low levels of inflation a country heavily involved in trade will favor a fixed exchange rate regime, but it may prefer more flexibility when the inflation rate is high and when a fixed exchange rate would lead to a loss of competitiveness. To test this hypothesis, total trade (exports plus imports) as a share of GDP is included in the regression as a separate variable and interacted with inflation.

The authorities of countries saddled with large external debt tend to be wary of devaluation of their currency, as it would increase the debt burden expressed in domestic currency. On the other hand, servicing the debt requires earning foreign exchange, which is easier when the economy is more competitive. Given this trade-off, it is difficult to tell exactly how high external debt ratios would influence the "preferred" level of the exchange rate. What seems to happen in practice is that the first argument prevails for some time, as policymakers cling to the old exchange rate for fear of creating balance sheet problems, but eventually balance-of-payments difficulties arise, and the currency is allowed to depreciate. It should also be noticed that large step devaluations may occur without a change in the exchange rate regime (e.g., under an adjustable peg). Alternatively, devaluations are frequently associated with changes in the exchange regime, but not with a particular direction of the change. Indeed, countries commonly fix the exchange rate at a level more depreciated than the pre-existing floating rate, while a transition from a peg to a float often occurs under speculative pressure, which leads to depreciation as soon as the currency is floated. Because of this, we cannot predict the direction in which external indebtedness would influence the exchange-rate regime, but we include the ratio of external debt to GDP in the analysis to investigate the matter empirically.

Tables I and II give summary statistics for the sample for the baseline regression. The data for all non-categorical variables come from the International Financial Statistics of the IMF. As we can see, all three exchange rate regimes distinguished by the IMF are almost equally represented in the sample, with independent floating occurring somewhat less frequently than the two other options. The mean inflation rate in the sample is 68.1 percent per year, which is quite high. At the same time, the median inflation in the sample is much lower, 22.5 percent. The mean is drawn to the right by relatively rare occurrences of very high inflation. The countries in the region trade a great deal with the outside world. There is considerable variation within the sample in the level of international reserves, which ranges from very low to quite high magnitudes. So does external debt. The unemployment rate also varies substantially.

The ordered logit regression results are summarized in Table III. Our baseline regression (Column 1) includes inflation, inflation squared, and dummies for lagged regime. There is a considerable degree of stickiness in exchange rate regime choice, as indicated by a large and highly statistically significant coefficient on the lagged dependent variable.

The idea of a nonlinear relationship between the inflation rate and the degree of exchange rate flexibility is confirmed by the regression. Both coefficients have the predicted signs and are highly statistically significant. The pseudo R-squared of 56 percent is fairly decent for such a parsimonious specification. (13)

The partial relationship between inflation and the degree of exchange rate flexibility, as given by the baseline regression, peaks at the inflation rate of 300 percent per year. When inflation is lower, the relationship is positive: the probability of choosing a fixed regime diminishes, and the probability of choosing an independent float rises if the rate of inflation increases, ceteris paribus. In this range, external competitiveness is the dominant concern. Above 300 percent per year, more inflation is more likely to be followed by a tightening of the exchange regime. Internal stability comes to center stage, and the exchange rate is used as a nominal anchor.

Table IV shows how the regime would be chosen if inflation and the previous regime were its only determinants. At low inflation the regime is likely to remain unchanged, (14) while at very high inflation a peg becomes the preferred alternative. While these outcomes comply with the model, the cutoff points are higher than what we had expected. At the same time, this table traces quite well the experiences of high inflation countries, as demonstrated by Table V. (15) The table accurately predicts Romania's drift from a peg to a managed float and to an independent float, Lithuania staying with its independent float, Poland staying with a peg, and Bulgaria switching to a fixed regime after its hyperinflation. In the cases of Albania in 1991 and Bulgaria in 1993, the model predicts a move from a fixed regime to managed floating, while the countries actually moved all the way to independent floating; these are the only discrepancies between predicted and actual outcomes. Still, it must be acknowledged that, while the model describes reasonably well, the behavior of the economies to the right of the "hump" of the regression curve, the experiences of the bulk of the economies are in the area to the left of the hump. In that sense one may say that the existence of concavity in the relationship between exchange rate flexibility and inflation is more firmly established by the evidence presented here than the fact that the slope of the relationship actually turns from positive to negative in the empirically relevant range.

When added to the baseline, the other control variables enter the regression with expected signs. The coefficient on the ratio of international reserves to M2 is negative and statistically significant. Economies with higher unemployment choose to have more flexible regimes, but the quantitative effect is quite small and statistically significant only at the 20 percent level. Openness to trade does not seem to be a major factor affecting the choice of exchange rate regime. The coefficient on trade is negative, and the coefficient on the interaction term between trade and inflation is positive, which is consistent with the model, but neither coefficient is statistically significant. The coefficient on the ratio of external debt to GDP turns out positive and statistically significant. (16) Apparently, countries with a high level of external indebtedness opt for flexible regimes, which tend to be associated with greater external competitiveness and facilitate earning foreign exchange necessary for servicing the debt. Throughout these exercises, the coefficients on previous regime dummies, inflation, and inflation squared change fairly little and retain high statistical significance. (17)

A potential problem with these regressions is endogeneity of the explanatory variables. In particular, inflation can certainly be affected by the choice of the exchange rate regime; this is both common wisdom and an empirical regularity, (18) as well as an important premise of the model. In the model, the exchange regime is determined on the basis of what inflation was in the past, and in the data the realization of inflation, calculated on the basis of average prices in a given year and in the preceding year, antedates the realization of the regime, taken at the beginning of the next year. This timing solves the problem of causality. The persistence of both the regressors and the dependent variable does not necessarily imply a bias in the results, since we explicitly control for the persistence of the regime in the regressions by including a lagged value of the dependent variable. As long as the functional form is correct and the disturbances themselves are not serially correlated, the regressors (lagged regime and inflation) will not be correlated with the disturbances, so the timing resolves the problem of simultaneity. Still, it would be comforting to find an appropriate instrument for inflation that would be highly correlated with the inflation rate and arguably uncorrelated with the future choice of the regime.

As noted in the introduction, all transition economies in Central and Eastern Europe experienced major inflationary shocks when they liberalized their price systems. We use a change in the index of liberalization of internal markets developed by de Melo et al. [1996]. The index reflects the authors' judgment, informed by country reports and expert opinions, on the extent of liberalization of domestic prices and abolition of state trading monopolies. It ranges from zero to one, with a 0.1 gradation, and increases with liberalization. The index is available for all the countries in our sample for the years 1989-94. (19) As Column 1 of Table VI indicates, the baseline regression in this subsample looks similar to the one for the entire sample (Cf. Table III, Column 1). (20)

Our measure of liberalization is simply the difference in the liberalization index between two consecutive years. A sharper change in this index would have a more dramatic impact on prices, so this change should be correlated with inflation. Indeed, the correlation coefficient is 0.52. To be sure, price liberalization is not the only reason for inflation, but it is an important one. On the other hand, the decision to liberalize prices is presumably independent of the exchange rate regime. Hence, this variable is a promising instrument. As the regressions include both inflation and its square, we need two instruments. We obtain a second instrument simply by squaring the change in the liberalization index.

Instrument variable estimation results are presented in the second column of Table VI. Compared with the simple regression, we lose some explanatory power. All the coefficients still have the expected signs and remain statistically significant. The absolute values of the coefficients on the two inflation terms approximately treble. We take this as a confirmation of our story. When only the exogenous components of inflation are considered, the reaction to inflation looks even sharper.

Table VII contains the results of fixed effects estimation. One might posit that each country has inherent characteristics that barely move over time. These might determine the choice of regime, and all the action we get in our regression comes from the lagged regime term, which captures these characteristics. If this were true, all the explanatory variables would have been rendered insignificant by an inclusion of country fixed effects. In fact, the nonlinear relationship between inflation and the choice of regime is still there when country fixed effects are included (Column 2). We get the same story from intra-country temporal variation that we get from the full panel. Controlling for year fixed effects (Column 3) does not change the picture either. This means that the assumption that disturbances are uncorrelated across countries in a given year is not crucial. Adding both country and year fixed effects (Column 4) still does not disturb the main message of the model.

Finally, we estimate the relationships above using our own codification of the degree of exchange rate flexibility rather than relying on the IMF's classification. This scheme has seven gradations, which are, in the order of increasing flexibility: a currency board, a conventional peg, a crawling peg, a horizontal band, a crawling band, a managed float, and an independent float. We assigned regimes to individual observations on the basis of the IMF's reports on Exchange Arrangements and Exchange Restrictions as well as Country Reports and Country Profiles by the Economist Intelligence Unit. The Pearson correlation coefficient between this and the IMF's measures of exchange rate flexibility is 0.81, and the Spearman rank correlation coefficient is 0.85.

Not surprisingly, the results are quite close to those obtained using the IMF classification. These regressions appear in Table VIII. The story on the importance of previous regime and inflation for the choice of the regime holds in this specification. The availability of foreign exchange reserves and the rate of unemployment influence the choice in the predicted way, but the former is only marginally significant and the latter is statistically insignificant. The level of trade openness does not seem to affect the choice of the regime.

The main message that we take from this regression exercise is that the nominal anchor versus the competitiveness tradeoff can be traced quite well in the choice of exchange-rate regimes in transition economies. In addition, exchange-rate regimes are shown to be sticky. This fact is obvious, but it is usually implicitly attributed to the stickiness of the variables underlying the choice, and the inclusion of a lagged dependent variable on the right-hand-side (motivated by institutional costs of adjustment) is rare in the literature. (21)

Other controls, such as foreign exchange reserves and the rate of unemployment, behave mostly in the predicted way. Countries with less international reserves are less likely to peg, as it is more difficult for them to peg. Countries with high unemployment want more room for expansionary policies and are less willing to sacrifice flexibility for low inflation. Openness to trade does not seem to affect the choice once we control for lagged regime and inflation.

VI. Conclusion

In this article we have developed a model of exchange rate regime determination which emphasizes the tradeoff between the use of the exchange rate to promote external competitiveness and its use to promote domestic price stability. This tradeoff is relevant for transition economies which have been subject to numerous inflationary shocks. The main implication of the model is that the relationship between the rate of inflation and the degree of exchange rate flexibility is expected to be nonlinear. A rise in inflation from a low level would call for more flexible arrangement. On the other hand, an increase in already high inflation would call for fixing the exchange rate.

I have tested this hypothesis on a sample of 13 transition economies for the years 1990-98. Ordered logit regressions confirm the main implication: the terms on inflation and its square have predicted signs and are highly statistically significant. The model has also passed a number of robustness checks, such as instrumental variable estimation, inclusion of country and year dummies, addition of other explanatory variables, truncation of the sample, and estimation using alternative coding for the dependent variable. After controlling for inflation, exchange rate regimes have been found to be highly persistent. The propensity to fix the exchange rate is weakened by low levels of international reserves with respect to broad money and by high unemployment.

While the model demonstrates the working of the exchange rate regime tradeoff, and the empirical results are supportive, the next challenge is to incorporate fiscal policy in the framework for analyzing inflation and exchange rate policy.

Appendix

Brief History of Exchange Arrangements in Central and
Eastern Europe (1990-1998)

Country Brief History

Albania The exchange rate for the lek is determined on the
 basis of supply and demand for foreign exchange.
Bulgaria The lev became internally convertible in 1991.
 A currency board was introduced in July 1997,
 pegging the lev to the DM.
Croatia The exchange rate of the kuna is determined in the
 interbank foreign exchange market. The National Bank
 of Croatia may set intervention exchange rates to
 level undue fluctuations in the exchange rate.
Czech Republic The koruna was pegged to a USD/DM combination in a
 narrow band in 1993.
 The band was broadened to 7.5% in 1996.
 The koruna was devalued and floated in 1997.
Estonia The kroon has been pegged to the DM in a currency
 board system since 1992.
Hungary The central rate of the forint was formally pegged
 to a basket (whose composition varied) in 1989, but
 5% fluctuations were allowed and frequent
 compensatory devaluations were taking place. A policy
 of preannounced monthly devaluations has been
 followed since May 1995.
Latvia Latvia gradually passed from the Russian ruble
 through the Latvian rublis to the Latvian lats,
 which became the sole legal tender in October 1993.
 Since February 1994, the lats has been pegged
 informally to the SDR.
Lithuania Similar transition from the ruble through the talonas
 to the litas. In April 1994 a currency board was
 introduced, pegging the litas to the USD.
Macedonia The National Bank of Macedonia participates in the
 wholesale foreign exchange market to maintain the
 value of the denar against the DM at a level that
 would meet balance of payments objectives.
Poland The zloty was pegged to the USD in January 1990. In
 May 1991, the zloty was devalued and pegged to a
 basket. In October 1991, a preannounced crawling peg
 was introduced, with occasional step devaluations.
 The fluctuation margin was widened to 2% in March
 1995, 7% in May 1995, and 10% in February 1998.
Romania The exchange rate for the leu was unified in
 November 1991. Still, the rationing of hard currency
 occurs and controls are applied to the markets for
 foreign exchange. Full internal convertibility of
 the leu was introduced on January 30, 1998.
Slovakia From July 1994 to October 1998 the koruna was pegged
 to a combination of the USD and the DM. The initial
 fluctuation band of 1.5% was widened to 3% in
 January 1996, 5% in July 1996, and 7% in January
 1997. On October 2, 1998 the currency basket and
 the fluctuation band were abolished.
Slovenia The tolar has been floating since its introduction in
 October 1991. The central bank interferes in the
 foreign exchange market with the objective of
 stabilizing the real exchange rate.

Table Ia

Exchange Rate Regime--IMF Classification

O-fixed; 1-managed floating; 2-independently floating

Regime Frequency Percent

 0 30 34.88
 1 33 38.37
 2 23 26.74

Total 86 100.00

Table Ib

Exchange Rate Regime--Author's Classification

0-currency board; 1 -peg; 2-crawling peg; 3-band; 4-crawling band;
5-managed float; 6-independent float

Flex Frequency Percent

 0 13 15.12
 1 16 18.60
 2 8 9.30
 3 3 3.49
 4 4 4.65
 5 20 23.26
 6 22 25.58

Total 86 100.00

Table II

Continuous Variables--Summary Statistics

Variable Obs Mean Std. Dev. Min Max

inflation 86 68.1 143.8 0.50 1082.60
trade_ratio 60 0.924 0.349 0.327 1.679
unempl 57 9.26 3.56 3.0 16.5
res_m2 81 0.348 0.189 0.035 1.235
debt_ratio 86 0.432 0.315 0.045 1.605

inflation--year-on-year percentage increase in the CPI

trade_ratio--(exports+imports/GDP)

unempl--year average unemployment rate, percent

res_m2--international reserves / M2 , end of year

debt_ratio--external debt/GDP

Table III

Ordered Logit Estimation

Dependent variable--degree of exchange rate flexibility according to
the IMF

Regressor 1 2 3

Reg_pr=1 dummy 4.19 4.83 3.90
 (5.10) (4.92) (3.81)
Reg_pr=2 dummy 6.67 7.37 6.36
 (6.21) (5.91) (4.43)
Inflation 0.0440 0.0329 0.0499
 (3.97) (2.88) (3.52)
Inflation Squared -0.0000733 -0.0000545 -0.0000830
 (-3.37) (-2.46) (-3.03)
Foreign Reserves -- -3.81 --
 (-2.12)
Unemployment -- -- 0.173
 (1.32)
Trade Ratio -- -- --
Trade*Inflation -- -- --
Debt Ratio -- -- --
Number of Observ 86 81 57
Pseudo R-squared 0.56 0.59 0.60

Regressor 4 5

Reg_pr=1 dummy 4.42 5.21
 (3.98) (5.09)
Reg_pr=2 dummy 7.94 8.69
 (4.53) (5.59)
Inflation 0.0506 0.0390
 (1.72) (3.70)
Inflation Squared -0.0000954 -0.0000722
 (-2.61) (-3.25)
Foreign Reserves -- --
Unemployment -- --
Trade Ratio -0.842 --
 (-0.49)
Trade*Inflation 0.00923 --
 (0.37)
Debt Ratio -- 5.02
 (2.93)
Number of Observ 60 86
Pseudo R-squared 0.67 0.63

z-statistics in parentheses

Table IV

Ordered Logit Estimation--Predicted Distribution

Regime--degree of exchange rate flexibility according to the IMF

 Previous Regime

New Fixed
Regime

Fixed [pi]<69.2; [pi]>531.6
MF 69.2<[pi]<531.6
IF --

 Previous Regime

New Managed Float Independent Float
Regime

Fixed [pi]>633.0 >680.5
MF [pi]<83.7; 517.1<[pi]<633.0 [pi]<16.1; 584.7<[pi]<680.5
IF 83.7<[pi]<517.1 16.1<[pi]<584.7

Table V

Outcomes for High Inflation Countries

Regime--degree of exchange rate flexibility according to the IMF

 Regime

Country Year Inflation Previous Predicted Actual

Romania 1992 211 1 2 2
Albania 1992 226 0 1 2
Romania 1991 231 0 1 1
Romania 1993 255 2 2 2
Bulgaria 1991 339 0 1 2
Lithuania 1993 410 2 2 2
Poland 1990 555 0 0 0
Bulgaria 1997 1083 2 0 0

Table VI

Instrumental Variable Estimation

Dependent variable--degree of exchange rate flexibility according to
the IMF Instruments--change in price liberalization index, square of
change in price liberalization index

Sample--1990-94

 Regressor 1 2
 Ordered Logit IV

 Previous Regime = MF 3.84 3.74
 (2.66) (2.83)
 Previous Regime = IF 5.05 6.19
 (3.41) (4.05)
 Inflation 0.0474 0.1444
 (3.20) (2.03)
 Inflation Squared -0.0000786 -0.0002507
 (-2.79) (-1.91)
Number of Observations 34 34
 Pseudo R-squared 0.55 0.47

z-statistics in parentheses

Table VII

Fixed Effect Estimation

Dependent variable--degree of exchange rate flexibility according
to the IMF

 regressor 1 2
 Baseline Country FE

 previous regime = MF 4.19 627
 (5.10) (3.24)
 previous regime = IF 6.67 9.30
 (6.21) (3.93)
 inflation 0.0440 0.0736
 (3.97) (3.65)
 inflation squared -0.0000733 -0.000119
 (-3.37) (-3.44)
number of observations 86 86
 Pseudo R-squared 0.56 0.71

 regressor 3 4
 Year FE Country and
 Year FE

 previous regime = MF 4.39 6.24
 (4.95) (2.90)
 previous regime = IF 7.55 10.06
 (6.02) (3.67)
 inflation 0.0425 0.0784
 (2.91) (3.09)
 inflation squared -0.0000713 -0.0001285
 (-2.51) (-2.79)
number of observations 86 86
 Pseudo R-squared 0.59 0.73

z-statistics in parentheses

Table VIII

Ordered Logit Estimation

Dependent variable--author's classification of exchange rate regimes.
Regressors include dummies for categories: peg; CP--crawling peg;
HB--horizontal band; CB--crawling band; MF--managed float;
IF--independent float. Omitted category--currency board.

regressor 1 2 3 4

Prev. Regime=peg 4.97 4.08 5.77 4.28
 (3.77) (3.00) (2.91) (3.09)
Prev. Regime=CP 6.38 5.49 7.68 5.37
 (4.37) (3.35) (3.20) (3.34)
Prev. Regime=HB 9.13 8.24 10.79 8.10
 (4.97) (4.31) (4.61) (3.98)
Prev. Regime=DB 7.97 7.53 9.65 7.29
 (4.80) (4.47) (3.65) (4.07)
Prev. Regime=MF 9.62 9.76 11.05 8.04
 (6.06) (5.90) (5.07) (4.62)
Prev. Regime=IF 13.3 13.55 15.36 12.15
 (7.31) (6.82) (5.09) (5.57)
Inflation 0.0238 0.0145 0.0243 0.0224
 (2.10) (2.15) (1.57) (1.63)
Inflation Squared -0.0000413 -0.0000253 -0.0000507 -0.0000376
 (-1.87) (-2.53) (-2.35) (-1.44)
Foreign Reserves -- -2.87 -- --
 (-1.58)
Trade Ratio -- -- 0.112 --
 (0.06)
Trade*Inflation -- -- 0.0123 --
 (0.70)
Unemployment -- -- -- 0.0908
 (0.86)
Number of Observ 86 81 60 57
Pseudo R-squared 0.49 0.53 0.53 0.45

z-statistics in parentheses

Notes

(1.) Unlike the previous factors, which are essentially transitional and almost exogenous, fiscal laxity reflects the policy stance. As such, it may be the source of persistent inflationary pressure, particularly if the deficits are financed by borrowing from the banking system. If that is the case, the choice of the nominal anchor is largely irrelevant, as no anchor can be consistent with fiscal profligacy, except in the very short run. The focus of this article is on fighting transitional inflation in the presence of inertial elements. Extending the model to a situation where the government keeps adding fuel to the inflationary fire is a matter for future research.

(2.) Of course, output declined in all CEE countries in early years of transition regardless of the choice of exchange rate regime. There are numerous "structural" reasons for this "transition recession." Still, a liquidity squeeze imposed by a money-based stabilization would smother production further. Fischer et al (1996) find a positive impact of a fixed exchange rate regime on real GDP growth in transition economies.

(3.) These models take root in the closed-economy literature looking at optimal conduct of monetary policy in the face of a trade-off between inflation and unemployment (e.g., Barro and Gordon [1983]).

(4.) Uncertainty in these models comes from terms of trade shocks.

(5.) The shortness of policymaker horizons is posited because of the difficulty of political and economic forecasting far into the future during transition and because of the fairly short average tenure of governments in the countries of Central and Eastern Europe (see, e.g., EBRD [1999, Chart 5.6]).

(6.) I am following Edwards in identifying the real exchange rate with the product wage in the tradable sector (or, equivalently, with the wage in dollars). Nothing of substance would change if I defined it as the ratio of nontradable to tradable prices. With this definition, the rate of real appreciation would equal

[s.sub.t] = [[pi].sub.Nt] - [[pi].sub.Tt] = [gamma](1 - [phi]) x [epsilon] / ([eta] + [epsilon][alpha] + [epsilon](1 - [alpha]) [gamma] [[pi].sub.t-1]

which is nearly identical to equation (10).

(7.) One might argue that as the countries of Central and Eastern Europe are catching up in terms of income with the western world, their currencies should experience equilibrium real appreciation, and s*> might be a reasonable target. It should be noted, however, that the loss function is attached to a model of the economy which does not incorporate such equilibrium real appreciation. One could allow for such appreciation by assuming [z.sub.t] to be positive in equation (4'). In that case inflation and real appreciation would be greater than the magnitudes given by equations (7') and (10), respectively. Hence, s, given by (10), can be seen as real appreciation on top of the equilibrium one.

(8.) d[[phi].sub.t] / d[[pi].sub.t-1] = [kappa] - 2[[zeta] - [theta]) + ([xi] + [theta]) [[phi].sub.t-1] - [kappa]([xi] + [theta])[[pi].sup.2.sub.t-1] / [[1 + ([xi] + [theta])[[pi].sup.2.sub.t-1]].sup.2]

This expression is positive when [[pi].sub.t-1] is small, negative when [[pi].sub.t-1] is large, and it has one positive root.

(9.) If [[phi].sub.t-1] + [kappa][[pi].sub.t-1] - [zeta][[pi].sup.2.sub.t-1] > 1 + [xi][[pi].sup.2.sub.t-1] then higher e will lead to less exchange rate flexibility.

(10.) When inflation is above zero but below [kappa]/([xi] + [zeta], the desired degree of exchange rate flexibility, [phi], is greater than 1, so it is the constraint 0 [is less than or equal to] [phi] [is less than or equal to] 1 that makes this whole segment a locus of possible steady states. If this constraint is removed, possible steady states are the whole horizontal axes (zero inflation) plus the point [phi] (= 1, [pi] = [kappa]([xi] + [zeta]). The path toward the latter steady state may look like a spiral.

(11.) For most authors the problem is just the opposite--the available characterization of exchange rate regimes is too detailed for them, and they have to decide how to aggregate across categories in order to create a dichotomous measure for testing their theories.

(12.) Currency boards have been established in Estonia (1992), Lithuania (1994), Bulgaria (1997), and Bosnia (1998). Lithuania pegged its currency to the US dollar, while the three other countries pegged to the deutsche mark.

(13.) Adding the two inflation terms to a regression of regime on the two previous regime dummies increases the pseudo R-squared by 17 percentage points and log likelihood by 16 units.

(14.) The right-hand side of the regression equation at zero inflation is just below the high cutoff point (and well within estimation errors) when the previous regime is an independent float, so 16.1 in the bottom right cell of the table is statistically indistinguishable from zero.

(15.) Arbitrarily defined here as countries having inflation over 200 percent per annum.

(16.) This is true regardless of the exchange rate used for conversion (end-of-period or period average). The result does not change if the ratio of external debt to exports is used.

(17.) Change is more pronounced when trade is included in the regression because of the interaction term.

(18.) See, for example, Fischer et al. [1996] or Ghosh et al. [1997].

(19.) Since 1994, a similar index has been produced by the European Bank for Reconstruction and Development and published in its Transition Reports. The index has remained virtually unchanged since 1994.

(20.) This truncation of the sample is an additional robustness check. We have also experimented with excluding certain countries from the sample, such as Bulgaria which had very high inflation in 1997 and appears to be an outlier, the Baltic states which were part of the former Soviet Union, and Albania, Croatia, and Macedonia where quality of the data might be low. The results were essentially the same as in the full sample.

(21.) I know of only one article, Bernhard and Leblang [1999], where this is done.

References

Alesina, Alberto and Allan Drazen, "Why Are Stabilizations Delayed?," American Economic Review, 81, 1991, pp. 1170-88.

Barro, Robert J. and David B. Gordon, "A Positive Theory of Monetary Policy in a Natural Rate Model," Journal of Political Economy, 91, 1983, pp. 589-610.

Barth, Richard C. and Chorng-Huey Wong, (eds.), Approaches to Exchange Rate Policy Choices for Developing and Transitional Economies, Washington, DC: International Monetary Fund, 1994.

Begg, David, "Disinflation in Central and Eastern Europe: The Experience to Date," in Cottarelli, Carlo and Gyorgy Szapary (eds.), Moderate Inflation: The Experience of Transitional Economies, Washington, DC: International Monetary Fund, 1998.

Bernhard, William and David Leblang, "Democratic Institutions and Exchange-rate Commitments," International Organization, 53, 1999.

Bruno, Michael, "Stabilization and the Macroeconomics of Transition: How Different is Eastern Europe?," Journal of Economics of Transition, 1, 1993, pp. 5-19.

Calvo, Guillermo A. and Carlos A. Vegh, "Inflation Stabilization and BOP Crises in Developing Countries," NBER Working Paper No. 6925, 1999.

Collins, Susan, "On Becoming Flexible: Exchange Rate Regimes in Latin America and the Caribbean," Journal of Development Economics, 41, 1996, pp. 117-138.

Coorey, Sharmini, Mauro Mecagni and Erik Offerdal, "Disinflation in Transition Economies: The Role of Relative Price Adjustment," in Cottarelli, Carlo and Gyorgy Szapary, (eds.), Moderate Inflation: The Experience of Transitional Economies, Washington, DC: International Monetary Fund, 1998.

Devarajan, Shantayanan, and Dard Rodrik, "Do the Benefits of Fixed Exchange Rates Outweigh Their Costs? The CFA Zone in Africa," in Ian Goldin and L. Alan Winters, (eds.), Open Economies: Structural Adjustment and Agriculture, Cambridge, UK: Cambridge University Press, 1992.

De Melo, Martha, Cevdet Denizer and Alan Gelb, "From Plan to Market: Patterns of Transition" Working Paper No. 1564, Policy Research Department, World Bank, 1996.

Devereux, Michael and Charles Engel, "Fixed vs. Floating Exchange Rates: How Price Setting Affects the Optimal Choice of Exchange-Rate Regime," NBER Working Paper No. 6857, 1998.

EBRD, Transition Report 1999: Ten Years of Transition, London, UK: EBRD, 1999.

Edwards, Sebastian, "Exchange Rates as NominalAnchors," Weltwirschaftliches Archiv, 129, 1993.

Edwards, Sebastian, "The Determinants of the Choice Between Fixed and Flexible Exchange-Rate Regimes," NBER Working Paper No. 5756, 1996.

Fernandez, Raquel and Dani Rodrik, "Resistance to Reform: Status Quo Bias in the Presence of Individual-Specific Uncertainty," American Economic Review, 81, 1991, pp. 1146-55.

Fisher, Stanley, Ratna Sahay, and Carlos Vegh, "Stabilization and Growth in Transition Economies," Journal of Economic Perspectives, 10, 1996, pp. 45-66.

Frieden, Jeffry, Piero Ghezzi, and Ernesto Stein, "Politics and Exchange Rates: A Cross-Country Approach to Latin America," IADB Research Network Working Paper No. R-421, 2000.

Ghosh, Atish R., Anne-Marie Guide, Jonathan D. Ostry, and Holger C. Wolf, "Does the Nominal Exchange Rate Regime Matter?" NBER Working Paper No. 5874, 1997.

Klyuev, Vladimir, Essays on Monetary and Exchange Rate Policy in Transition Economies, Doctoral Thesis, Harvard University, 2000.

Koch, Elmar B., "Exchange Rates and Monetary Policy in Central Europe--a Survey of Some Issues," Working Paper No. 24, Austrian National Bank, 1997.

Masson, Paul R., "Monetary and Exchange Rate Policy of Transition Economies of Central and Eastern Europe after the Launch of EMU," Policy Discussion Paper No. 99/5, International Monetary Fund, 1999.

McKinnon, Ronald D., "Optimal Currency Areas" American Economic Review, 53, 1963, pp. 717-725.

Mundell, Robert A., "A Theory of Optimal Currency Areas," American Economic Review, 51, 1961, pp. 657-665.

Rodrik, Dani, "The Rush to Free Trade in the Developing World: Why So Late? Why Now? Will It Last?" in Stephan Haggard and Steven B. Webb, (eds.), Voting for Reform: Democracy, Political Liberalization, and Economic Adjustment, New York, NY: Oxford University Press, 1994.

Sachs, Jeffrey D., "Economic Transition and the Exchange-Rate Regime," American Economic Review: Papers and Proceedings, 56, 1996, pp. 147-152.

Sweeney, Richard. J., Clas G. Wihlborg, and Thomas D. Willett, (eds.), Exchange-Rate Policies for Emerging Market Economies, Boulder, CO: Westview Press, 1999.

Tavlas, George, "The New Theory of Optimal Currency Areas" World Economy, 16, 1993, pp. 663-85.

Tornell, Aaron and Andr6s Velasco, "Money-Based versus Exchange-Rate-Based Stabilization with Exogenous Fiscal Policy," NBER Working Paper No. 5300, 1995.

Tomell, Aaron and Andres Velasco, "Fiscal Discipline and the Choice of Nominal Anchor in Stabilization," Journal of International Economics, 46, 1998, pp. 1-30.

Turnovsky, Stephen J., "The Relative Stability of Alternative Exchange Rate Systems in the Presence of Random Disturbances" Journal of Money, Credit and Banking, 8, 1976, pp. 29-50.

Williamson, John, (ed.), Currency Convertibility in Eastern Europe, Washington, DC: Institute for International Economics, 1991.

Vladimir Klyuev
International Monetary Fund