Competitiveness in Transition Economies: What Scope for Real Appreciation?
Publication Date:
01 Jan 1998
eISBN:
9781451974164
Language:
English
Keywords:
SP; paper number; EU country; monetary policy; transition economy; equilibrium wage; labor force; private sector; current account; Real exchange rates; Productivity; Competition; Human capital; Baltics; Central and Eastern Europe
Download PDF (4 MB)

We estimate equilibrium dollar wages for 15 transition economies of Central and Eastern Europe (CEE) and the former Soviet Union. Equilibrium dollar wages are Interpreted as full employment wages consistent with a country’s physical and human capital endowment, and estimated by regressing actual dollar wages on productivity and human capital proxies in a short (1990- 95) panel of 85 countries. The main results are: (1) equilibrium dollar wages have appreciated steadily in the Baltic countries and fast-reforming CEE transition economies, but have been fiat in most CIS countries: and (2) 1996 actual dollar wages remain below estimated equilibrium dollar wages for most but not ail transition countries covered.

Abstract

We estimate equilibrium dollar wages for 15 transition economies of Central and Eastern Europe (CEE) and the former Soviet Union. Equilibrium dollar wages are Interpreted as full employment wages consistent with a country’s physical and human capital endowment, and estimated by regressing actual dollar wages on productivity and human capital proxies in a short (1990- 95) panel of 85 countries. The main results are: (1) equilibrium dollar wages have appreciated steadily in the Baltic countries and fast-reforming CEE transition economies, but have been fiat in most CIS countries: and (2) 1996 actual dollar wages remain below estimated equilibrium dollar wages for most but not ail transition countries covered.

IN THE LAST few years, the currencies of the transition countries of Central and Eastern Europe (CEE) and the former Soviet Union have undergone large real appreciations (Figures 1 and 2).1According to a widespread consensus, the process of appreciation set in following an initially strongly undervalued position of these currencies.2 Accordingly, it was not perceived to be threatening to the competitiveness positions of these countries. In view of the magnitude of real appreciations witnessed—up to fifteen-fold in some CIS economies, including Russia, since early 1992—this can no longer be taken for granted. The Czech currency crisis in May 1997, which led to a devaluation of the koruna by 8.5 percent following a large widening of the current account deficit in 1996 (Table 1), has focused new attention on the question of whether the scope for real appreciation has by now been exhausted in some or most transition economies. If the answer is yes. this would clearly have implications for the design of exchange rate policy, which in many countries so far has been guided primarily by the objective of reducing inflation, rather than maintaining competitiveness.

Figure 1.
Figure 1.

Real Exchange Rate Indices Versus the U.S. Dollar: Centeral and Eastern Europe

Citation: IMF Staff Papers 1998, 003; 10.5089/9781451974164.024.A004

Figure 2.
Figure 2.

Real Exchange Rate Indices Versus the U.S. Dollar: Baltic and CIS Countries

Citation: IMF Staff Papers 1998, 003; 10.5089/9781451974164.024.A004

Table 1.

Current Account Balances

(In percent of GDP)

article image
Source: International Monetary Fund.

The objective of this paper is to obtain a sense of what scope, if any, remains for real appreciation in transition economies of Central and Eastern Europe and the former Soviet Union before competitiveness becomes an issue. At the outset, one might wonder whether the recent current account positions in these countries offer any information in this regard. While some countries have been in surplus—particularly energy exporters—most were in deficit (Table 1). But the latter is just what standard intertemporal equilibrium models would predict for countries that are rebuilding their capital slocks following a large structural shift. More generally, at a time when these economies are adapting to large relative price shocks, including terms of trade shocks, major trade disruptions, and other institutional changes affecting the trade regime, any inference about the appropriateness of the real exchange rate based onthecurrent account position seems even more difficult than usual.3

What methodology could be usefully employed to decide whether the currencies of transition economies should by now be considered overvalued? The literature on competitiveness and real equilibrium exchange rates suggests a variety of approaches. Competitiveness in a given year is sometimes assessed by comparing the value of a real effective exchange rate (REER) index in that year with its value in some reference year in which the economy is regarded as being in external and internal equilibrium.4Alternatively, an equilibrium real exchange rate can be estimated, and compared to the actual real exchange rate. This is typically done either by first estimating or assuming an “equilibrium current account” and then estimating the real exchange rate that would generate it. or by estimating a reduced-form model in which the equilibrium real exchange rate is identified with the long-run real exchange rate that is associated with steady-state net foreign assets (NFA) and current account positions.5

The first of these approaches is difficult to apply to transition economies since it implicitly assumes a constant real equilibrium exchange rate over time and as such does not take account of changes in productivity, capital stocks, tastes, or commodity prices, which would, in general, imply a change in the equilibrium rale. Such changes presumably matter in the context of transition economies undergoing rapid structural transformation. Even more important, the REER index approach requires a reference year for which equilibrium of the real exchange rate can be assumed. No year before the beginning of transition can be taken as a reference, since trade and capital flows in that period were heavily restricted. On the other hand, the initial period of external opening and reform is usually associated with very large real depreciaiions. In general, one would thus assume that tran-sition economies have gone from artificially appreciated currencies—such as the official ruble exchange rate of 1.7 to the dollar in 1991—to undervalued currencies following currency convertibility and the initial floating or pegging of the exchange rate. There is no discernible state of equilibrium between these two states that could serve as a reference.

Unfortunately, the alternative and more sophisticated econometric approaches described above are not feasible either. The joint estimation of long-run equilibrium exchange rates and the current account or NFA positions is precluded by the absence of adequate time-series data for these countries, with only 2 to 5 years of data since the beginning of the transition process. The two-step approach, on the other hand, requires the estimation of real exchange rate elasticities of the current account. Even ignoring the fact that this estimation typically occurs in a time-series context and is thus subject to the same data limitations, it would seem difficult, if not impossible, to estimate the effects of real exchange rate movements on the current account at a time when fluctuations in exports are likely to be driven primarily by such changes as the removal of export quotas, the breakdown of traditional trading blocks, and changes in relative prices within the tradables sector.

The solution we propose to overcome these problems rests on two ideas. First, we use a real exchange rate measure that is both readily available for transition economies and—unlike REER indices—can be directly interpreted and compared in levels, namely, dollar wages in the manufacturing sector. Second, we estimate equilibrium dollar wages as a function of productivity measures using a short panel of countries, rather than a’into series. Thus, the estimated equilibrium dollar wage represents an estimate of what the country could “afford” based on its stock of human and physical capital. We then go on to interpret competitiveness as the gap between actual dollar wages and estimated equilibrium dollar wages.

The approach pursued is inspired by the way in which macroeconomic practitioners often form a judgment of the international competitiveness status of a country—by comparing the country’s average dollar wage with that of other countries that are considered “similar” in terms of the remaining determinants of profitability or unit cost, such as the quality and quantity of human and physical capital. Which countries are to be considered “similar” in this sense is usually decided ad hoc. We put this informal comparison on a more systematic footing by constructing a fictitious country with identical human and physical capital to the country we are assessing—as measured by crude proxies, which will be discussed in detail below—and estimating the dollar wage one would expect to prevail in this country.

Since one of the productivity measures on the right-hand side of our wage equation is (PPP-adjusted) GDP per employee (or per member of the labor force, or per capita), our approach encompasses competitiveness comparisons based on aggregate unit labor cost (see Havlik. 1996, who compares unit labor costs in Central and Eastern European transition economies with that of Austria). It is also related to a large literature comparing international price levels and relating deviations from purchasing power parities lo differences in resource endowments across economies.6 Unlike this literature, however, our left-hand side variable is dollar wages, not prices or price indices. Apart from the desire to stay close to the terms in which the discussion on competitiveness is led by practitioners, there are two reasons for this. First, the link from structural determinants to national price levels usually runs via factor prices, the argument typically being that factor prices are an important determinant of the price of nontradables relative to tradables, which in turn is an important determinant of the price level.7 As a result, specifying a country-invariant structural equation relating resource endowments to price levels that can be estimated from observable variables seems even harder than specifying a corresponding regression model of dollar wages. Second, even if we wanted to estimate real equilibrium exchange rates for transition economies based on a comparison of national price levels, we would have been constrained by data availability, as the International Comparison Program, on which price level comparisons are typically based, was not extended to all of these countries at the time of this analysis.

Finally, we owe much to a recent paper by Halpem and Wyplosz (1997), who pursue broadly similar objectives for a somewhat different set of tran-sition economies that does not include the Baltic and CIS economies studied here (except for Russia).8 The basic similarity between Halpern and Wyplosz (1997) and this paper is the attempt to estimate equilibrium dollar wages in transition economies using a set of productivity proxies as right-hand side variables. However, the estimation approaches and data sets are different. Halpern and Wyplosz argue that to uncover the relationship between fundamentals and equilibrium dollar wages one needs to “observe each country for a long period of time” and consequently use a long panel (1970-90), which includes the planned economies (with a planned economy dummy) and a time trend. We argue that, since we are interested in equilibrium dollar wages as opposed to long-run steady-state wages, it is enough to use a cross section (or alternatively a short panel to test for country specific effects), provided the countries in our sample have been market economies for a sufficiently long time that we can assume that on average they are in equilibrium. On this basis, we use a short panel (1990-95) that includes 70 market economies and 15 transition economies with a transition dummy to capture out-of-equilibrium effects for this group (see Table A1). Apart from allowing us to estimate equilibrium wages for the Baltic and CIS countries, for which long panel data do not exist, this enables us to estimate equilibrium wages for transition economies within sample (1991 to 1995) rather than out of sample, as is the case for Halpern and Wyplosz. In spite of these differences, the main results of the two papers are fairly close for the countries they both study. Both papers suggest that while the gap between actual and equilibrium dollar wages declined in most countries since the beginning of transition, it remained substantial in 1995-96 in most (but not all) cases. For example, for Russia. Halpern and Wyplosz estimate equilibrium dollar wages of 400-500 dollars for 1995. Our estimate is lower (235-394 dollars, depending on the specification and data used), but this is still substantially higher than actual dollar wages in manufacturing during 1995 and 1996 (107 and 188 dollars, respectively).

I. Methodology

Basic Approach

In attempting to estimate equilibrium dollar wages for transition economies, we face two difficulties. First, as emphasized in the introduction, the virtual absence of time-series data for these economies; and second, the need to make inferences about equilibrium wages based on the observation of actual wages, which might be far off equilibrium. Our proposed solution to this problem is to estimate equilibrium dollar wages using a cross section (or alternatively to deal with country-specific fixed effects short panel) of countries, including many nontransilion economies. This is briefly justified as follows.

Suppose country i is a market economy. By this we mean that market forces have been determining prices and wages in country i for a long period of time (say. ten years). Suppose we knew the equilibrium dollar wage of country i. Then, assuming we have no further information about countryi. our best guess of actual dollar wages in this country would be the equilibrium dollar wage:

E[w$,i|w$,i*]=w$,i*.                         (1)

Now suppose we do not know w*s, j but we have a theory of how it is determined

w$,i*=G(Zi),                       (2)

where Zj denotes a vector of observable “fundamentals” in country i. If we can also observe the actual dollar wages for each country. Gcould beestimated. Combining equations (1) and (2). we obtain

w$,i*=G(Zi)+εi                       (3)

with E[εi╿Z] = 0. Thus, equilibrium dollar wages could be estimated by regressing actual dollar wages on Zi in a cross section of countries for which equation (1) holds and computing the fitted wage for each country.

Three issues remain to be addressed. First, the above assumes that we know the modelG(Zi).We thus need to propose such a model. Second, we need to find observable proxies forZi Finally, we need to decide how to treat observations from transition economies, for which equation (1) cannot be assumed. The next section deals with the first of these issues, while the other two will be taken up in the sections that follow.

A Model of Equilibrium Dollar Wages

Short-Run Equilibrium

Consider a simple two-sector model of equilibrium dollar wages and the real exchange rate. We make two key assumptions. First, the tradables sector is assumed capital intensive relative to the nontradables sector. This is necessary and sufficient to generate a positive relationship betweentherelative price of nontradables and dollar wages and thus justify our use of dollar wages as a measure of the real exchange rate. For simplicity, take the special case in which we have a Ricardian technology in the nontradables sector:

YN=δLN                           (4)

and a Cobb-Douglas technology in the tradables sector:

YT=ΓKΑLT1-Α                      (5)

Assuming that the international (i.e.. dollar) price of tradables is given, so that the domestic price of tradables equals the exchange rate (Pr = E, whereEis defined as domestic currency per dollar), equilibrium in the nontradables sector then implies a linear relationship between the real exchange rate and dollar wages:

PN=Wδ;orPNPT≡P=1δWE≡1δWS.                           (6)

Our second key assumption is to assume a cost to capital adjustment, so that capital can be treated as given in the short run. This enables us to separate short-run equilibrium—that is. equilibrium dollar wages conditional on the level of the capital stock, which is all we need to estimate equation (3) in a cross section—from the dynamics of capital accumulation and long-run steady state. Tradables profit maximization leads to

WE≡W$=Γ(1-;Α)(KLT)Α                                  (7)

Assume now that labor is in fixed domestic supplyĽ and mobile across the two sectors. Using the fact that all nontradables production is necessarily consumed at home, labor market clearing implies

W$=γ(1-Α)(KL¯-1δCN)Α                                  (8)

To close the model, we need to make some assumption about the con-sumption side. The simplest possibility is lo assume that capital is owned by foreign investors and domestic workers-consumers are concerned with static utility maximization only, that is. they do not save and they only worry about how to allocate each period’s wage bill across the two sectors. Assuming Cobb-Douglas preferences with expenditure sharesβ for nontradables and l-β for tradables, this leads to

CN=βWL¯PN=βδL¯                     (9)

Substituting in equation (8) then gives us an expression for wage deter-mination in general equilibrium:

W$=γ(1-α)(K(1-α)L¯)α.                    (10)

In its log-linear version, this equation says that equilibrium dollar wages depend on a constant (which might be different across countries if technology is different) and the aggregate capital-labor ratio. The elasticity of dollar wages with respect to capital is the (constant) capital share α.

The price of simplicity in this way of closing the model is that—because we are not allowing consumers to accumulate debt or save—we are imposing current account balance. This is probably too strong: most economists would agree that the notion of “external equilibrium” defining the equilibrium dollar wage need not require current account balance at all times. Instead, it only requires the current account deficit or surplus to be consistent with an intertemporally optimizing, sustainable consumption path. In the simplest intertemporal extension of the case considered before, consumers solve

maxΣt=0+11+pIn(CN,tβCT.t1-β)s.tΣt=0+11+r(CN,tPt+CT,t-Ws,tL¯)=0.                     (11)

The proportion of nontradables and tradables in infratemporal consump-tion is given (as before), as CN, i=β/ (1-β) CT, i/Pi In the simplest case, whenr = pand the intertemporal consumption profile is flat.9 equation (8) becomes

Ws=γ(1-α)[KL¯-β1-βC¯T/Ws¯]α.                    (12)

As in equation (8), the equilibrium dollar wage will depend on technol-ogy, capital and the labor force, but the simple linear dependence of Ws on the log ofK / Ldisappears. This is because labor is no longer allocated across tradables and nontradables sectors in a fixed proportion. if the capital stock increases, dollar wages and nontradables prices rise, nontradables consumption declines and a larger proportion of the labor force moves lo the tradables sector. This generates an offsetting effect that dampens the appreciation of Ws.10

Dynamics and Long-Run Steady State

Equations (10) and (12) define relationships between equilibrium wages and fundamentals along the lines of equation (2), which can in principle be estimated using actual dollar wage data, as argued above. However, before turning to issues of empirical implementation it is instructive to take the model one step further to see what it implies aboul the dynamics of real exchange rates in transition economies. Note that nothing in our empirical approach, which we return to below, depends on the particular view taken in this subsection.

Equations (10) and (12) suggest that the dynamics of the equilibrium dollar wage will be determined by differential productivity improvements in the tradables and nontradables sector and capital accumulation in the tradables sector. If one abstracts from the former for the time being (i.e., treats the technology parameters as fixed), combining the previous section’s assumptions about production and consumption with a standard neoclassical growth model will thus deliver real exchange rate dynamics.11 Capital will be accumulated as long as the dollar profit from installing an extra unit of capital—which, inter alia, depends on the prevailing dollar wage—exceeds the unit installation cost times the international interest rate. As capital is installed, the marginal product of labor rises and equilibrium dollar wages increase. Thus, the adjustment of equilibrium dollar wages to their steady-state level can be depicted as a movement along the curve defined by equations (10) or (12) (see Figure 3). The steady-state level of equilibrium dollar wages itself will depend on the technology parameters of the model, the international interest rate, and installation costs. In particular, higher productivity in the tradables sector will imply higher steady state dollar wages.

Figure 3.
Figure 3.

Example of Dollar Wage Dynamics with Inflation Inertia

Citation: IMF Staff Papers 1998, 003; 10.5089/9781451974164.024.A004

From equation (6), Figure 3 equivalently traces out the dynamics of the real equilibrium exchange rate for given productivity parameters. In steady state, the “pure” Balassa-Samuelson mechanism reemerges: long-term trends in the real exchange rate will be driven by differential rates of (total factor) productivity growth in the tradables and nontradables sectors; faster productivity growth in the tradables sector generates a real appreciation. During the adjustment to steady state, however, there is an additional force behind real appreciation, namely capital accumulation in the tradables sector. If productivity of the service sectors rises faster than tradables productivity during the adjustment to steady state, we have two opposing forces acting on the real exchange rate. Real appreciation will still result if capital accumulation causes dollar wages to outpace relative productivity gains by the nontradables sector. Equilibrium dollar wages, on the other hand, will unambiguously rise during adjustment as long as productivity in the tradables sector does not decline.

The discussion so far has focused on equilibrium real exchange rates and dollar wages because it was based on a market clearing, fully optimizing current account model. However, the equilibrium model may not provide a good description ofactualshort run dollar wage and real exchange rate movements for well-known reasons. For example, in a flexible exchange regime, exchange rates might be driven by external borrowing and portfolio investment in addition to capital investment; this could generate swings in the nominal exchange rate that, in the presence of short-run wage rigidities, will feed through to dollar wages. In a fixed exchange rate regime, on the other hand, the real exchange rate may become misaligned if there is inflation inertia, that is, if price or wage growth depends to some extent on past price or wage growth. In this case, wages could exhibit dampened oscillations around the equilibrium adjustment path, with unemployment arising whenever actual dollar wages are above their equilibrium levels for any given level of the capital-labor ratio (see Figure 4).12

Figure 4.
Figure 4.

Example of Dollar Wage Dynamics with Inflation Inertia

Citation: IMF Staff Papers 1998, 003; 10.5089/9781451974164.024.A004

The initial overshooting of the equilibrium dollar wage path will occur either if there is some inflation in the system to begin with, or—even in the absence of initial inflation—if the dollar wage was initially undervalued, as shown in Figure 4. At least one of these conditions is likely to apply in transition economies that fix exchange rates at the beginning of transition.

In summary, the basic stylized fact documented at the beginning of the paper—a sharp real appreciation since the inception of transition in practically all transition economies—could be interpreted as follows. At the beginning of the transition, real exchange rates in transition economies are (1) below their steady-state levels and(2) undervalued, that is, below their equilibrium levels conditioning on existing levels of profitable capital (point WS0 in Figure 4). The former is a result of the capital obsolescence effect associated with external opening and price liberalization: the latter can be thought of either as consequence of an initial monetary overhang or initial capital (light/capital outflows that are not captured by the intertemporal current account model. As transition proceeds, new capital is accumulated in the tradables sectors, leading to an appreciation ol equilibrium dollar wages and real exchange rates. Actual real exchange rales appreciate in the direction of this moving target, but in the presence of inflation inertia or capital-account driven appreciation there is no guarantee that this process will stop once equilibrium real exchange rale/equilibrium dollar wage levels have been reached.“Several years into transition, this raises the question whether dollar wages have by now overshot equilibrium or not.

Empirical Specification

We now return to the problem of estimating equilibrium dollar wages on the basis of observable economic fundamentals at any given point in time. Equation (10) or (12) suggests that equilibrium dollar wages should depend on the capital share and equilibrium (i.e.. full employment) productivity in the tradables sector, which in turn depends on total factor productivity in the tradables sector, the (equilibrium) capital labor ratio, consumption preferences, and possibly wealth (as a determinant of the consumption level). On this basis, it should be possible to run a regression along the lines of equation (3). Before we do so. we need lo address some problems ol empirical implementation13.

Measuring Productivity

The first challenge is to find a set of observable right-hand side variables consistent with equations (10) or (12). Whether we pick measures that proxy individual variables or parameters on the right-hand side of equations (10) and (12) (such as total factor productivity and the capital labor ratio) or capture a combination of variables (such as tradables productivity) is a matter of empirical convenience, as we are not trying to isolatetheeffect of individual “fundamentals” on equilibrium wages. The approach we take follows Halpern and Wyplosz (1997) in using a wide set of relatively crude productivity measures or determinants, namely. (1) normalized PPP-adjusted GDP as a broad productivity measure, (2) a schooling or human capital variable, (3) the share of agriculture in GDP as a general proxy for economic development, and (4) a dummy for OECD membership, also as a proxy for economic development.14 In addition, we try lo include various indicators capturing institutional factors that potentially influence productivity (such as property rights). We justify this procedure as our best hope of proxying for productivity fundamentals in the tradables sector in view of CI) the unavailability of sector-specific productivity data for the tradables sector in our countries; and (2) the need to proxyequilibrium(i.e., full employment) productivity on the basis of observable measures, as follows.

Suppose we had data on equilibrium output in the tradables sectors across countries, valued at a set of average international prices. For simplicity, assume further that in equilibrium there is an unknown fixed ratio between the sizes of the tradables and nontradables sector. For example, take the case of equation (10). Denoting the average international price of tradables by ITwe would have:

TTYT*=ITΓKαLT*(1-α)                       (13)
LT*=(1-β)L¯                          (14)

Then, dividing equilibrium output al international prices by the labor force would result in a perfect measure of tradables productivity:

ITYT*L¯=ITγ(K(1-β)L¯)α≡gdpT*.             (15)

From equation (10),-(I-a) / lT -gdp*s . .Thus, if equation (I) holds, running a regression of actual wages on gilp*T in a cross section or panel of countries will be sufficient to estimate equilibrium dollar wages. Note that this procedure would only assume thatais the same across countries: pref-erences and all remaining technology parameters could differ.

In fact, gdp*[ is unavailable, but we do have PPP-adjusted GDP, that is.aggregateGDP at international prices.1,1 Abstracting from the problem that this may not be measured at labor market equilibrium, dividing PPP-adjusted equilibrium GDP by the labor force yields

gdpT*≡ITYT*+INYN*L¯=gdpT*+INδβ                 (16)

where IN denotes the international average price of nontradables. Using equation (10), we obtain:

WS*=(α-1)+INδβIT(1-α)ITgdp*                      (17)

It follows that a regression of actual dollar wages ongpd* and a constant will only lead to consistent estimates of equilibrium wages under implausibly strong assumptions, such as cross-country equality not just of α but also of β and δ that is, preferences and nontradables productivity. However, not only are β and δ likely to differ across our sample, but they are probably correlated withgpd*(e.g., both tradables and nontradables productivity might depend on the stage of economic development). Moreover, PPP-GDP is not generally measured at full capacity (or equilibrium) but based on actual output and thus incorporates some degree of cyclical variation.

With these problems in mind, we adopt the Halpern-Wyplosz approach outlined at the beginning of this subsection. To the extent that cross-country differences in preferences and nontradables productivity are correlated with differences in economic development and economic and political institutions, additional regressors reflecting these differences should help control for differences inβ and δ. This motivates the inclusion of variables such as human capital, agriculture share. OECD membership, and institutional indicators. On the one hand, these are likely to be directly related to productivity. On the other hand, they may also contain information aboul the relative size of nontradables consumption15.

Finally, to minimize the possible distortions arising from cyclical variation ingpd* we explore the implications of different normalizations of PPP-adjusted GDP. using not just the labor force but also population and total employment in the denominator. The justification for this procedure will be briefly discussed in Section II beiow.

Country-Specific Effects

Based on the discussion so far, the error term in a regression of dollar wages on the variables suggested above may contain a country specific component. First, we might obviously still miss some country-varying determinants of equilibrium wages. In this case, running OLS on a cross section could generate a misspecitication problem as these country varying determinants will probably be correlated with some of the right-hand side variables.16 Second, even assuming that our right-hand side variables fully account tor equilibrium wages, disequilibrium models along the lines of the example illustrated in Figure 4 suggest that the extent of disequilibrium at any given point in time could depend on the country’s initial our-of-equi-librium position (or on large shocks or regime changes that have placed the country far out of equilibrium later in its history). In general, we have no information on this initial out-of-equilibrium position (the major exception being the transition economies, see below), which may or may not be correlated with the country’s “fundamentals.” If it is, we are back to the fixed-effects problem described previously. If it is not, E[ws, i w*s, i]= w*s, i will still hold—in other words, the country-specific effects will be random—and OLS will give us unbiased and consistent estimates. However, allowing for a non-time-varying country-specific component in the error term may still be a good idea from the point of view of efficiency.

To address the potential presence of country-specific effects, we run our regression on a short panel rather than a cross section. Apart from increasing our data set. this allows us to test for the presence of fixed effects and random effects and apply the appropriate estimator. The tests will be discussed below.

Transition Dummies

Recall the argument underlying assumption (1), namely, that countries have undergone a sufficiently long adjustment period in a regime governed by market forces to make it a priori impossible to guess whether their wages are undervalued or overvalued. This assumption is clearly violated for transition economies, which have only recently become subject to market forces, and whose dollar wages might have been far out of equilibrium at least in the early transition years (the conventional wisdom being that they were highly undervalued). Thus, if we include the transition economies in the regression sample without any further ado, our equilibrium wage estimates forgiven levels of the right-hand side measures are likely to be biased downward. To avoid this, one could either leave the transition economies out of the regression sample altogether, or include them in the sample but add a “transition dummy” to the regression. The latter procedure is more efficient, since it exploits information from the additional countries to estimate the slope parameters of the regression. Note that the transition dummy needs to be time varying, since one cannot assume that the extent to which the transition countries are out of equilibrium stays the same over the period covered by our panel. Similarly, it would seem advisable to include separate dummies for Central and Eastern Europe and countries of the former Soviet Union, since the former began their transition earlier than the latter.

In summary, our basic regression model is

w$i,t=a0+a1argi,t+a2gdpi,t+a3schooli,t+a4ocedi+ΣΤ=1TaΤ+4ceei,t+ΣΤ=1TaΤ+9fusi,t+μi+εi,t,    (18)

where wsi, t stands for (he log of dollar wages,agri, tfor the log of the percentage share of agriculture in GDP,gdpi, tfor the log of normalized PPP-adjusted GDP,schooli, tfor a human capital proxy based on secondary school enrollment data, which are discussed below,oecdτ i, tfor a dummy that equals 1 ifi was an OECD member over the entire sample period and 0 otherwise, and ceeτi,1 for a dummy variable that takes on the value I in period /1 ifi =τ and if countryi is a transition economy of Central or Eastern Europe and 0 otherwise. fsui, t is an analogous dummy for the CIS and Baltic countries and p, is a country-specific effect. Expressing wsi,1, gdpi,1, andagri,1in logs follows convention and contributes toward normality of the error in equation (18) in view of the skewness and nonnegativity of the wage and income distributions. The model was also extended to control for differences in taxation, the degree of government intervention, and property rights (see Section II).

We are left with a final issue, namely, what to do about the transition dummies when computing the fitted wages based on equation (18) for transition economies, which in accordance with equation (1) will be interpreted as equilibrium wage estimates. Clearly, it would be wrong to include the transition dummies when we compute fitted wages, since these dummies will pick up the potentially large initial exchange rate misalignment that might be common to transition economies—this is why they were added. Thus, comparing actual dollar wages in a transition economy with its fitted dollar wage including the transition dummy will tell us whether this economy’s dollar wage is under- or overvalued relative to the average under- or overvaluation in the group of transition countries, not relative to the equilibrium dollar wage corresponding to its human and physical capital endowment. The relevant comparison is thus between the actual dollar wage in a transition economy and its fitted wage based on equation (18) after setting the transition dummies to zero. The implicit assumption is that in all aspects not captured by the three time-varying right-hand side variables (and later on. the additional variables introduced to account for institutional differences), the transition economy is structurally similar to an average non-OECD economy in our sample. Put differently, we need to assume that the transition dummies included in the regression reflectonly the average extent of exchange rate misalignment in transition economies during the sample period, rather than structural differences between transition economies and nontransition developing countries. We return to this assumption below.

II. Estimation and Testing

Data

Our data set begins in 1990. the earliest starting date or’economic transition in Eastern Europe, as defined by the first comprehensive attempt in a formerly planned economy to both liberalize prices and open the economy.17 The endpoint was chosen to be 1995, the last year for which a large cross-sectional coverage could be achieved. Between these two dates, we attempted to include all market economies for which data were available for any number of consecutive years, and all transition economies for which a continuous sequence of annual data points was available from their first year of transition onward. This led to an unbalanced panel of 85 countries, including 15 transition economies: Bulgaria, the Czech Republic, Hungary, Poland. Romania, the Slovak Republic, the three Baltic states. Belarus. Kazakhstan, the Kyrgyz Republic, Moldova. Russia, and Ukraine. Our starling dates are 1990 for Hungary and Poland. 1991 for the remaining Eastern European countries, and 1992 for the Baltic and CIS countries. The market economies in our sample include all OECD countries, most Latin American countries, and some African and Asian economies.18

The economic variables used in our regressions were constructed as follows.

  • Dollar wagesare average monthly wages in manufacturing in U.S. dollars. Data on manufacturing wages in national currencies were obtained from ILO publications and from the OECD “Short Term Indicators Transition Economies” database. In addition, we used national statistical publications and. in some cases, the IMF’s “Recent Economic Developments” country reports to fill in gaps and to broaden the coverage of the sample. In cases when information on hours was not available, the hourly wage data were converted into monthly wages by assuming an 8-hour working day and a 4.3-week month. Monthly wages were then expressed in U.S. dollars using annual average exchange rates from the 1FS. In order to ensure cross-country comparability, we made every effort to obtain wage data for employees for each country. In some cases, however, only wage rates for workers were available, which lend to be substantially lower than employee wages. We included these countries in the sample but attempted to control for the difference in the definition ofthedependent variable by including an appropriately defined dummy variable on the right-hand side.19

  • Data for purchttsing-power-parity-adjusted GDP(“PPP GDP” for short) were obtained from the IMF’s World Economic Outlook (WEO) database. The purchasing power parity estimates used by the WEO are based on the Penn World Table Mark 5 and (for transition countries) on a comparative study by the United Nations Economic Commission for Europe, and extended using “bridging equations”; see Wagner (1995). The prob-lem with these data is lhal the estimates for transition countries are based on a pretransition production structure that is bound to become increasingly inaccurate as transition proceeds. The World Bank and the European Union (EU) provide cstimaies based on more recent price comparisons, but for the transition economies they are only available for one year, 1995, and in the case of the EU data, only for the Eastern European economies. Consequently, we generally relied on the WEO data for the purposes of estimation but used World Bank and EU data to check the robustness of our equilibrium wage estimates (see below).20

We experimented with normalizing PPP GDP by total population, labor force, and employment (obtained from the WEO database and from the World Bank’s Social Indicators of Development database) to construct three alternative proxies for overall productivity. The argument presented in Section I was based on normalization by the labor force: however, this assumed that PPP GDP could be measured at its equilibrium (i.e., full capacity) level. In the presence of unemployment or overemployment, normalization by actual employment might be preferable, as this is likely to imply the least cyclical productivity measure. However, this measure has the problem that it will, all things being equal, overstate productivily in economies with a large subsistence sector. Normalizing by the labor force or, alternatively, population (as in Halpern and Wyplosz and most of the literature on explaining deviations from PPP) will avoid that particular problem. In the end, we decided lo run our regressions using all three measures to check the robustness of the results to the choice of normalization. Normalized PPP GDP is denoted by “gdp”.

  • The share of agriculture inGDP (“agr”)was taken from the World Bank’s 1977 Social Indicators of Development“(SID) database. Since the SID database did not contain 1995 shares of agriculture, we decided to use the lagged, rather than contemporaneous, agriculture share in our regressions. This procedure can be justified on the grounds that the importance of agricultural production over the short time period considered is likely to capture cross-country differences in technological development rather than time trends; hence, the lagged share of agriculture is a satisfactory proxy to use.21 A few missing values for 1994 and earlier years were filled in using national statistics or were interpolated.

  • Ahuman capital variable (“school”)was constructed from the SID database as a measure of the average level of secondary school education of the labor force in each country and year. More precisely. school i, t is the average of secondary school enrollment ratios in country i between 1950 (the first year reported in the SID) and t-1, weighting the enrollments ratios in each year with the relative size of the cohort of fifteen-year-olds al the time. Since the first year reported in SID for school enrollment is 1965. we assumed 1965 schooling levels for the earliest cohorts. We also assumed lower participation rates for these cohorts in our sample period.22 While this method is obviously crude, we consider the resulting indicator a better proxy for the human capital stock than either contemporaneous or lagged data on schooling for a particular cohort. In addition to the human capital variable thus constructed, we also experimented with alternative measures (average years of primary, secondary, and higher education in the population in the 15+age group; share of people with different levels of education in the 15+ age group) from a database compiled by Barro and Lee (1996).

  • To control for possible cross-country differences in the legal framework of economic activity, which are likely to influence productivity, we augmented the list of right-hand side variables byindicators of government intervention, tax structure, and property rights and an overall index of “economic freedom” constructed by the Heritage Foundation. The indicators are based on 1994-95 information.

Econometrics

As discussed in Section I, the main rationale for estimating equation (18) in a panel regression is to address the presence of country-specific effects in the determination of equilibrium wages.23 The simplest form of accounting for country-specific effects—that is, country-varying determinants of equilibrium wages that are not captured by the right-hand side variables in equation (18)—is to rewrite equation (18) as

w$i,t=a0+ΣΤ=13ajxji,t+a4ocedi+ΣΤ=1TaΤ+4ceei,t+ΣΤ=1TaΤ+9fusi,t+μi+εi,t,            (19)

where Xj, j= 1...3 denotes the economic determinants of equilibrium wages in equation (18) and μi stands for unmeasured country-specific effects. If μi is uncorrelated with the remaining right-hand side variables, its presence will merely generate a serial correlation problem that can be corrected by estimating equation (19) using a random effects (RE) estimator. If, on the other hand, μi is correlated with any of theXj we face a much more serious problem. On the one hand, the endogeneity of the error term induced byμiwill preclude consistent estimation of equation (19) using the pooled OLS or random effects estimators. On the other hand, first difference (FD) or fixed effects (FE) estimators, which are based on transforming equation (19) in a way that eliminatesμido not allow the estimation of thecee andfsu dummies and of the constant a0 and therefore preclude the computation of fitted wages for the transition economies (see Appendix for details).

Whether or not fixed effects are present, equation (19) may suffer from the standard endogeneity problem of a contemporaneous correlauon between the residual εi, t and one or more of the economic right-hand side variables. Since the agriculture share(agr)and the human capital indicator(school) are based on lagged GDP. agricultural production, schooling, and population growth, the most plausible candidate for this type of correlation is normalized PPP-GDP (gdp).24

To address these problems, we proceeded in two steps. First, we performed standard Hausman tests25 based on estimating equation (19) in first differences, that is, comparing the plain OLS FD estimates of equation (19) with FD estimates using lagged right-hand side variables as instruments. Lagged endogenous variables are clearly less than ideal as instruments, but unfortunately we had no better alternative. The FD specification is appropriate because it is immune to any additional endogeneity problem through the presence of fixed effects. The null hypothesis of no endogeneity cannot be rejected at conventional significance levels for all three normalizations, p-values for testing H0 “in the first-differenced version of equation (19). the residuals are orthogonal to the right-hand side variables” versus HA”residuals are correlated with the right-hand side variables” are repotted in the last row of the FD columns in Table 2. We thus conclude that an endogeneity bias through the contempora-neous correlation of ε and any xj is likely to be negligible.

Table 2.

Estimated Coefficients and Specification Test Results

(Dependent variable: monthly average wages in manufacturing, in U.S. dollars; standard errors are in parentheses)

article image
article image
Notes: Specifications (1), (2), and (3) refer to normalization of GDP by population, employment, and the labor force, respectively. Coefficients on cee and fsu dummies are not directly comparable across FD, FE, and RE estimates, see Appendix II. For FE and RE models, "test p-value" refers to a specification test against the closest model to the left. For FD models, "test p-value" refers to a specification test against FD models using lagged variables as instruments.

Based on these findings, we assume that weak exogeneity is satisfied and as a result the FD estimator is consistent. The next step is to test for the presence of fixed effects. There are several ways of accomplishing this. Since the FD estimator is consistent, we could perform a Hausman-type lest for the presence of fixed effects by comparing the FD estimates of the parameter vectorain equation (19) (consistent both underthenull of no fixed effects and under the alternative) with the pooled OLS estimates ofa(consistent only under the null of no fixed effects). Alternatively, we could perform the more conventional test for the presence of fixed effects by comparing the FE and the RE estimates ofa, but in this case we need to test for strict exogeneity first (Xji, t uncorrected with εi, t at all leads and lags) since without this property both FE and RE will be inconsistent whether or not fixed effects are present. This can be achieved via a Hausman-type test of FD versus FE (see Appendix). Yet alternatively, we could test FD against RE, in which case we would be jointly testing strict exogeneity and the absence of fixed effects. As it turns out, all three procedures imply that in our case neither the absence of fixed effects nor strict exogeneity can be rejected, and as a result it is legitimate to use the RE estimator.

In the following, we present the results of the procedure of sequentially testing strict exogeneity and the absence of fixed effects. The columns headed by FD and FE in Table 2 report the point estimates and standard errors forâFEand âFD, the parameter vector estimates obtained by the FE and FD estimators respectively. Table 2 also reports the rejection levels associated with the null hypothesis of strict exogeneity in the last row of columns FE; for details on how the test statistic underlying these p-values was computed, see the Appendix. As is apparent from the p-values reported. the null of strict exogeneity cannot be rejected at any conventional significance level.

We proceed to test H0: “individual-specific effects are orthogonal to the residual” (i.e.. absence of fixed effects but presence of random effects) versus HA :”individual-specific effects are correlated with the residual” (presence of fixed effects). Assuming strict exogeneity. FE is consistent under both, while RE is only consistent under the null. In addition, underH0 the RE estimator is (asymptotically) efficient—not just more efficient than FE—which implies that the hypothesis can be tested via a standard Hausman-test.26 Point estimates and standard errors for the (untransformed) âREare presented in Table 2 in the columns headed by RE. The p-values for the test procedures (reported in the last row) indicate that for all three specifications, the null of random effects cannot be rejected at conventional significance levels.

Table 2 also shows theestimated coefficientsfor each model specification. The RE estimates of all coefficients have the expected sign. In particular, normalized PPP GDP(gdp)is positive and highly significant across specifications. This remains true for both the FE and FD estimates. Based on the RE estimates, a higher share of agricullure(agr), which in a crosscountry setting is associated with a lower degree of development and thus lower productivity, results in lower dollar wages in all three specifications. The corresponding coefficient is significant at the 5 percent level in the specifications with PPP GDP normalized by employment and labor force, but significant only at the 15 percent level in the specification with PPP GDP per capita. The RE point estimate of the human capital variable(school)is positive in all specilications but it is significant only in the equation with PPP GDP normalized by employment. For both the agriculture share and human capital, the variation in the significance level of the estimated coefficient is due to differences in the magnitude of the coefficient. The OECD dummy(need) is positive and highly significant in ail specifications, its magnitude suggesting a sizable “dollar wage bonus” for OECD countries. The FE and FD estimates for the coefficients onagrandschoolare typically insignificant and in some cases have the wrong sign. Since most of the variation in the underlying variables is due to cross-country differences and not to variation over time, this result is not surprising.

The RE estimates show that the coefficients on the transition dummies(cee andfsu)are negative and significant throughout the sample period, suggesting lower than warranted dollar wages for these countries.27 The absolute value of the dummies shows a steady decline.28 This time path of the estimated dummies is consistent with the view that the transition economies were undervalued initially and that the gap has been closing over time.

Comparing the three specifications reveals differences in the estimated coefficients on the PPP GDP variables, agriculture share, human capital, and the OECD dummy. In particular, (i) the coefficientson gdpseem to be fairly close to one and to each other in the specifications which use labor force and population for normalization purposes, but only about half this size in specification (2), which uses employment; and (ii) in the presence of the employment-based productivity variable, the additional productivity or development proxiesagrandschoolare significant, whereas in the other two specifications they tend to be insignificant. One possible interpretation for the latter is that, controlling for population and labor force, employment may show swings across countries depending on the size of the subsistence sector; these swings are unrelated to productivity in the tradables sector and thus to wages but they are correlated withagrandschoolas indicators of development.

Note, finally, that the three specifications produce virtually identical values for the regressionR2, that is. the right-hand side variables account for the same percentage of variation in dollar wages in the three equations. This, and the fact that there are noticeable differences in the coefficients across specifications, suggest that we examine the implications of all three specifications on estimated equilibrium dollar wages (See next section).

To check therobustnessof the estimated coefficients we reestimated the equations with the following modifications. First, instead of using our human capital variable we used various combinations of the educational indicators by Barro and Lee (1996). Second, we restricted the sample to observations with average employee wages. Third, we experimented with using the share of mining in GDP as an additional regressor. Fourth, we attempted to account for cross-country differences in institutional factors influencing productivity by augmenting the list of variables by (I) an indicator of taxation; (2) an indicator of the extent of government intervention; (3) an indicator of the firmness of property rights; and (4) an overall index of economic freedom. Fifth, we experimented with alternative PPP GDP estimates for transition economies using EU and World Bank data. In this case, we used EU and World Bank data for the transition economies.29 but kept IMF WEO data for the remaining countries, as neither of the other two data sources provide panel data for all countries and years covered in the sample.

As it turns out, our results are not sensitive to any of these modifications.

On this basis, we decided to base our estimates of equilibrium dollar wages on the three parsimonious RE specifications reproduced in Table 2 and indicate the extent to which the estimated equilibrium wages are sensitive to the use of alternative PPP-adjusted GDP data (even if the regression coefficients are not, the fitted wages for individual transition economies may well be). For the purposes of deciding which currencies were undervalued in 1996 and which were not, we show estimates based on PPP-adjusted GDP data from all three sources.

Equilibrium Wage Estimates

Results

The RE estimates of the magnitude of the transition dummies suggest that dollar wages in countries of Central and Eastern Europe as a group were 75-65 percent under equilibrium (that is, their actual wages were at 25 to 35 percent of the warranted wage) in 1990 and 1991. This compares with 55 to 45 percent undervaluation in 1995. For the Baltics and the CIS countries, we see about 90 percent undervaluation in 1992 and about 60 percent in 1995. in conclusion, for our sample’s transition economiesas a group, though the gap between actual and equilibrium wages declined, it does not seem to have been eliminated by 1995.

We now examine whether this result holds for individual countries.Tables 3and 4 report actual and estimated equilibrium wages for Central and Eastern European and Baltic and CIS countries in our sample. Our equilibrium wage estimates are fitted wages based on the RE model, setting both the estimated country-specific effect and the transition dummies to zero,30 and using IMF WEO data on PPP-adjusted GDP per capita, for each of the three specifications. The sensitivity of these results to the use of PPP-adjusted GDP data from different sources is discussed at the end of the section. Note that the “ratio” columns of the tables give actual wages in percent of estimated equilibrium wages.

Table 3.

Actual and Estimated Dollar Wages in Manufacturing, US$/month: Selected Central and Eastern European Countries

(Standard errors are in parentheses)

article image
article image
Note: “ratio” denotes actual wage as a percentage of fitted wage.
Table 4.

Actual and Estimated Dollar Wages in, Manufacturing, US$/month: Baltic and Selected CIS Countries

(Standard errors are in parentheses)

article image
article image
Note: "ratio" denotes actual wage as a percentage of fitted wage.

The main points emerging from Tables 3and 4are as follows.

First, with minor exceptions,actualdollar wages exhibit a similarprofileacross countries, namely monotonically rising throughout the transition process, with particularly sharp increases in the first two years. However, dollar wagelevelsvary greatly across countries, and these differences are quite persistent through time. While it is possible to identify subgroups for which there is considerable convergence in actual dollar wages over the period (e.g., the Czech Republic, Estonia, Hungary, Poland, the Slovak Republic, and possibly Latvia), the dispersion within the transition country group as a whole does not seem to decline substantially during the period examined.

Second, there are interesting differences in the profiles of estimatedequilibriumdollar wages across countries. With the caveat that the standard errors around these estimates are fairly high,31 two broad patterns can be distinguished. In one group of countries—the Czech Republic. Hungary. Poland, the Baltics, and to a lesser extent. Bulgaria and Romania—estimated equilibrium wages increase more or less monotonously throughout the transition, typically after falling once at the beginning of our sample period. Except for the initial decline, which could be due to an overesti-mation of productivity in the early stages of transition, when the obsolescence of part of the capital stock has not yet been reflected in a decline in output, these countries broadly follow the pattern of continuous real equilibrium appreciation predicted hy the benchmark neoclassical model. In the CIS countries, on the other hand, we observe declining or tlat wages over the entire observation period. This is particularly true for Belarus, Kazakhstan, Moldova, and Ukraine, where equilibrium wages first decline and then stay flat, while in Russia and the Kyrgyz Republic equilibrium wages seem to turn the corner only in 1995. One interpretation might be that in these countries the process of new capital accumulation that drives recovery and real appreciation in transition countries according to the simple neoclassical model had not set in by 1995. for reasons that are not captured in that model, but that could relate to differences in the speed and consistency of reforms, the legal and political environment, and possibly location.32

Third, looking at actual and equilibrium dollar wages on a country-by-country basis confirms our earlier conclusion that despite their rapid increase, actual dollar wages did not catch up with equilibrium wages by 1995. The results in Tables 3and 4 suggest that at the outset of the transition process, dollar wages in manufacturing in the Czech Republic. Poland. Romania, and the Slovak Republic were about 30-4O percent of their equilibrium level, while dollar wages in Bulgaria, theBailies, and the CIS countries seem to have started out with an even larger discrepancy from their equilibrium level (10-20 percent). At the other extreme, our estimates put dollar wages in Hungary in 1990 at 50-60 percent of the warranted wage. This clearly indicates that—based upon our numerical estimates—all transition countries in our sample entered the transition with “too low” dollar wages (or, in our interpretation, with an undervalued exchange rate). By 1995, actual dollar wages in Central and Eastern European countries Mood between 35 and 75 percent of equilibrium wages, with Bulgaria al the low and Hungary and Poland at the high end of the range. The dispersion in the extent of undervaluation in 1995 is at least as pronounced for the Baltics and the CIS countries, where actual dollar wages range from 25-30 percent of estimated equilibrium wages (Belarus. Ukraine) to around 70 percent (Latvia, Lithuania, Kazakhstan). These results imply that for speeilications (1) and (2), the actual wage is outside the one-stan-dard-error band around the equilibrium wage estimate in all cases and the hypothesis that the equilibrium wage is smaller than the actual wage is for-mally rejected throughout at either the 1 or 5 percent level.33 According to specification (3). finally, the mill is rejected at the 1 or 5 percent levels for all countries except Kazakhstan. Latvia, and Lithuania in 1995 (/j-values 0.153. 0.059. and 0.051, respectively).

Using EU or World Bank data PPP-adjusted GDP data for transition economies does not affect any of these qualitative conclusions. In particular, the same observations apply regarding cross-country differences in the profiles of titled wages, and the overall conclusion that actual dollar wages remained below estimated equilibrium wages in 1995 and before. However, there are noteworthy differences in the equilibrium wage estimates for individual countries. In particular, using EU or World Bank data leads to equilibrium wage estimates for Estonia. Latvia, and Lithuania that are much closer together, implying that the hypothesis that the equilibrium wage is smaller than the actual wage could now be formally rejected in the case of Lithuania but perhaps not Estonia (depending on the specification). In addition, using World Bank data makes Kazakhstan look belter, increasing its equilibrium wage estimates by about 60 dollars.

In summary, even in countries where dollar wages rose sharply in the course of transition, they seem to remain below equilibrium, in most cases at less than 70 percent of the equilibrium estimate. Thus, we are confident that most transition countries did not have overvalued currencies by 1995. Unfortunately, owing to the unreliability of PPP-adjusted GDP data, there is considerable uncertainty in determining the set of countries that might constitute exceptions. We return to this thorny issue in the subsection below on competitiveness in 1996.

Discussion

Before proceeding, we now ask whether and to what extent the findings of the previous subsection could exaggerate the actual state of competitiveness in transition countries. In principle, we could have erred in two ways: equilibrium wages for transition economies could be overestimated, and reported wages in these countries could understate actual wages. We discuss each in turn.

Equilibrium wages for transition economies could be overestimated as a result of suppressing the transition dummies when computing fitted wages. As explained in Section I. transition dummies were included to capture the possibility that dollar wages in transition economies might be far off their equilibrium path, particulariy at the early stages of transition. The implicit assumption was that in any aspects relevant to equilibrium dollar wages not captured by the other right-hand side variables, transition economies were similar to a representative developing country during the period in question. If this assumption is violated, then any unrepresented structural idio-syneractes would be absorbed into the transition dummies. If the net effect of these factors is to depress equilibrium dollar wages in transition economics relative to nontransition developing countries, suppressing the transition dummy in the calculation of equilibrium dollar wages for transition economies would lead to overestimation34.

In our view, this problem could arise for the early years of transition, in particular because at this time PPP GDP per capita or worker could be a very poor proxy of productivity in the tradables sector. At the beginning of transition, measured real GDP has typically not yet declined lo reflect the obsolescence of large sectors of the economy, thus, PPP-adjusted GDP will overstate profitable productive capacity. However, once output begins bottoming out after several years of transition and real output declines on the order of 3050 percent, this effect is unlikely to be present; if anything, one would suspect that PPP-GDP per capita now understates actual productivity since much emerging private sector activity is unrecorded. Thus, from this angle, one would not expect 1995 equilibrium dollar wages to be overestimated. This still leaves the possibility that important determinants of equilibrium dollar wages—such as property rights, the legal framework, or political uncertainty—might be poorly captured by the productivity measures used. However, including direct measures of these variables in the regression did not seem to affect substantially the estimates of equilibrium wages, as discussed previously. Moreover, to introduce an upward bias in the equilibrium dollar wage estimates for transition economies, transition economies would need to do significantly worse in terms of unmeasured institutional conditions than the average developing country in our sample. This may be plausible for the beginning of transition, but not—in most cases—for the later years. On this basis, the equilibrium wage estimates for 1995/96 in countries such as Hungary, the Czech Republic, the Slovak Republic. Poland, Estonia. Latvia. Lithuania, and the Kyrgyz Republic is more likely to be biased downward than upward. In other cases—such as Belarus, which, in a sense, is still atthebeginning of transition—the possibility that equilibrium wages could indeed be overestimated seems more plausible.

We are left with the question whether the undervaluation of currencies in transition economies might be exaggerated due to the systematic underreporting of actual wages in transition economies, rather than the overestimation of equilibrium wages. There is good reason to suspect that wage underreporting could in fact be present. First, reported wages for transition economies could understate actual remuneration relative to nontransition economies because in-kind payments and other nonwage benefits may be a large share of earnings in transition economies, particularly at the beginning of the transition process.This may contribute toward explaining some of the unbelievably low actual dollar wages contained in Table 2 for the early transition years. Second, while this bias is likely to decline as transition progresses, it could be replaced by a second bias, namely, inadequate reporting of private sector wages. To the extent that newly emerging private firms are not fully captured, the official wage data reported for manufacturing or “industry” will mostly reflect public and formerly public (privatized) enterprises, whose behavior may be more similar to state-owned enterprises than to new private enterprises.35 If these enterprises pay lower wages than new private sector enterprises, this would imply lower reported wage levels not only relative to the true sector average but, more important, relative to the market wage level relevant for new entrants. Moreover, even if total remuneration in the public or formerly public sector and the new private sector is similar, public or formerly public firms might still pay a larger share in the form of noncash benefits, and thus reported wages would be higher if based on the new private sector.36

The evidence on public-private sector wage differentials, typically based on survey data, is mixed. For Poland (1993). Rutkowski (1994) finds a substantial earning differential for employees with post-secondary and university educations, who earn 22-26 percent more in the private sector, but not for lower education levels, where private sector jobs tend to pay somewhat less than public sector jobs. However, he also notes that “statisticians at CSO believe that earnings in the private sector are underreported to a larger degree than in the public sector” (p. 35). In a more recent paper. (Rutkowski, 1996), he finds 27-32 percent higher earnings at the post-secondary level, and 3-11 percent higher private earnings at the pre-secondary level of educational attainment. For the Czech Republic, Flanagan (1995) finds that after controlling for schooling and experience,”workers in new private firms earn about 18 percent more than workers in current or former state enterprises” in 1993. Using similar data but without controlling for schooling or experience. Vecerm’k (1995) reports wage differentials of 23 percent vis-à-vis the state sector and about 9 percent vis-à-vis the privatized sector for 1993 and about the same (24 and 10 percent, respectively) for 1994. This is roughly consistent with data reported in Ham. Svejnar, and Terrell (1995), who find that in the Czech Republic”private firms generally tend to pay slightly higher (0-10 percent) average wages than state enterprises” without distinguishing between privatized and new emerging private firms. On the other hand, for 1993, Blanchard, Commander, and Coricelli (1995) lind that private sector wages in Poland and Hungary are somewhat lower than public sector wages (by 7 and 14 percent, respectively), while they are somewhat higher in Bulgaria (16-50 percent) and much higher in Russia (82 percent). On Bulgaria. Beleva, Jaekman, and Nenova-Amar (1995) report that”private firms tend to pay higher wages, but this is to some extent offset by lower bonuses and more limited provision of nonwage benefits” (p. 219).

From Tables 3 and 4 it is clear that correcting for wage underreporting on the order of magnitude of the private-public wage differentials reported for each country would somewhat weaken, but not reverse our basic result. In other words, even if one takes the view that wages are undcrreported by the full extent of the private-public sector wage differential—which amounts to assuming that private sector wages do not enter the aggregate statistics at all—actual wages would still appear undervalued. This is even true for the outlier among the cases reported above, namely. Russia: applying the factor of 1.82 to the ratio of actual and equilibrium dollar wages inTable 3 still gives a ratio of only 0.73. Thus, in spite of the caveats discussed, it is hard to escape the overall conclusion that the currencies of the transition economies studied in this paper remain undervalued at least up lo 1995, substantially so in many cases.

A final issue is that actual dollar wages in transition economies may undcrrepresent wagecostsowing to differences in payroll taxes. From the firm perspective, payroll taxes are typically extremely high in economies at the early stages of transition, implying a downward bias in measured wage costs. Over time, this downward bias tends to become smaller as the payroll tax structure becomes more similar to that prevailing in market economies and nonwage benefits decline. As a result, our data may substantially overstate the gap between actual and equilibrium dollar wages in the initial years of transition, and create the illusion that this gap is closed fast.37 However, most of this effect will be concentrated in the first few years of reforms, and as such it should, in most cases, not constitute a proh-lem in assessing whether or not transition economies were wage competitive in the recent past. To the ex lent that specific countries are known to still have extraordinary high payroll taxes as of 1995 or 1996. a corresponding adjustment can be made individually.

Relative Competitiveness

The equilibrium wage argument pursued so far corresponds to a competitiveness concept that focuses on the attractiveness of a country to international capital flows. The implied perspective is that of an international investor, who compares all countries and invests in those where dollar wages are lowest relative to productivity, taking into account a convex adjustment cost. However, this may not be the right perspective if we lake the view that comparisonswithinthe transition group are, for any reason, more relevant than comparisons between transition economies and developing market economies, or if we believe that it is more relevant to compare dollarwages and productivity within groups of actual trading partners rather than globally across all potential competitors. In this case, the right question is not whether dollar wages in a given transition economy are below equilibrium wages for that transition economy but rather whether dollar wages arerelatively morebelow equilibrium in that economy than the dollar wages of its trading partners.

To answer this question, we constructed the following”index of relative competitiveness” for country i:

ui+Wi*WiΠj(Wi*Wj) Θi,j,                        (20)

where w*t / wt denotes the ratio of equilibrium and actual dollar wages for country i. θi, t denotes country i’s trade share with countryj so the denominator of equation (20) is a trade-weighted average of the equilibrium to actual dollar wage ratios of country i trading partners. In practice, we picked the six most important trading partners for each transition economy in our sample based on 1994 export and import data from the IMF’s Direction of Trade Statistics, and used 1994 trade weights throughout the sample period.38 The index μi is increasing in relative competitiveness; it takes a value smaller than one if countryidoes not have a cost-competitive edge over its trading partners, and greater than one if country i’s cost-com-petitive position is favorable.

For transition economies with a large share of trade with other transition economies (as is the case for most CIS countries), the index μi has the added advantage that it is less sensitive to omissions of transition-specific structural determinants of equilibrium wages, as such omissions would affect all transition economies in the same direction. In other words, we may believe that the simple comparison between actual and estimated equilibrium wages exaggerates competitiveness for some or all CIS economies because of unmeasured idiosyncracies of those economies that tend to depress the equilibrium wage, but this argument would not apply if a country is judged competitive on the grounds of the index μi. On the other hand, a disadvantage of usingμi relative to the earlier approach is thatμi is sensitive to mis-measurement of the actual wages among the country’s trading partners, rather than just mismeasurement of its own wage. For example, gross undcrrecording of the average wage of a major trading partner can render a country”uncompetitive” based on μi. even if its own wage is far from its correctly estimated equilibrium wage. In addition.μi will typically be less precise than point estimates of equilibrium dollar wages, as its construction involves the use of several (typically seven) such estimates.

Tables 5 and 6 show the result of our calculations for Central and Eastern European countries, the Baltics, and CIS countries in our sample. Our results show all Central European transition economies as competitive during the 1991-95 period, although Hungary’s competitiveness edge was never very large (and basically zero during 1993-94) and the Czech Republic’s edge has narrowed continuously since 1991 (Table 5). For the Baltics and the CIS countries, a more mixed impression emerges. While the competitiveness margins are generally larger than those of the CEE economies, three countries—Kazakhstan, Lithuania, and Latvia—appear less competitive than their trading partners by the end of the period (μi falls below the threshold value of one). For Latvia and Lithuania, this result is driven by fast dollar wage growth during 1993-95 relative to their main transition economy trading partners (especially Russia) and by the relatively low estimated equilibrium wage for Lithuania implied by IMF WEO estimates of PPP-adjusted GDP.39 For Kazakhstan, the driving force is that actual dollar wages were of roughly the same magnilude as Russia’s from about 1993 onward, while estimated equilibrium wages were substantially lower, Using PPP-adjusted GDP data from the EU and the World Bank changes these results in the case of Lithuania, whose relative competitiveness index rises above one (see below) but not lor Kazakhstan and Latvia. Moreover, while the alternative GDP dala imply a smaller estimated competitiveness margin for Estonia, this margin remains positive. Thus, the results of this subsection appear somewhat more robust to the use of PPP-adjusted GDP data from alternative sources than those of the previous subsection.

Table 5.

Competitive Position Relative to Trading Partners: Selected Central and Eastern European Countries

article image
Note: Indices prior to 1992 use 1992 information on Russia’s competitiveness.
Table 6.

Competitive Position Relative to Trading Partners: Baltic and Selected CIS Countries

article image

Competitiveness in 1996

We now assemble the main estimation results in this paper to evaluate the competitive position of transition economies using the most recent available annual wage and exchange rate data. In doing so, the main limitation is that we are forced to use equilibrium wage estimates out of sample. that is. we need to compare 1995 equilibrium wage estimates with 1996 dollar wage data. The justification is that, from Tables 3 and 4, by 1995 equilibrium dollar wages are either flat or rising in almost all countries. Thus, if actual dollar wages in 1996 continue to be significantly below 1995 estimated equilibrium wages (or the “relative competitiveness index” using 1995 equilibrium and 1996 actual wage data continues to be significantly larger than one), we can generally infer that the economy under discussion was indeed competitive in 1996.

Table 7 shows actual 1996 dollar wage data for our group of countries and compares it with 1995 estimated equilibrium wages using PPP-adjusted GDP data from the WEO. and a relative competitiveness index based on these estimates and 1996 actual wages, for each of the three RE specifications discussed in the previous subsections. Table 8 shows how these estimates would be affected if World Bank or EU GDP data were used instead of WEO data.40Thus, the tables capture the sensitivity of our results along two dimensions: (1) the choice of normalization of PPP-adjusted GDP: and (2) the source of PPP-adjusted GDP data. Information on competitiveness is provided from two perspectives. First, one can compare actual wages in transition economies with estimated equilibrium wages, which are the wages one would expect in an average developing market economy with identical right-hand side “fundamentals” as the transition economy we are studying. Since this average is taken over all 85 economies in our sample, the implicit assumption is that the transition economy of interest is in a potential competitive situation with all other economies that we use in our estimation. The second perspective is one where the gap between actual and estimated equilibrium dollar wages matters only relative to the corresponding trade-weighted gap between actual and equilibrium wages among the economy’s trade partners. This is expressed in the competitiveness index, which takes a value greater than one if the economy’s dollar wages are more undervalued (or less overvalued) than that of its trading partners. The implicit assumption is that the economy is in a competitive situation only with its trading partners: however, rather than directly comparing dollar wages within this group we compare “undervaluation gaps” to control for differences in fundamentals across countries.

Table 7.

Estimated Competitive Position in 1996 Using Alternative GDP Estimates: Selected Transition Economies

article image

Based on 1996 actual wages and 1995 estimated equilibrium wages.

Table 8.

Estimated Competitive Position in 1996 Using Alternative GDP Estimates: Selected Transition Economies

article image

Based on 1996 actual wages and 1995 estimated equilibrium wages. Ranges are based on results from different specifications.