As Germany has generally been considered the "anchor" of the system (see, e.g., Fratianni and von Hagen ... currency are
IS EUROPE AN OPTIMUM CURRENCY AREA ? BUSINESS CYCLES IN THE EU
Guglielmo Maria Caporale Centre for Economic Forecasting London Business School and Nikitas Pittis Department of Economics University of Cyprus Discussion Paper No. 15-98 July 1998
Financial support from ESRC grant no. R000237486, EMU and Financial Markets, is gratefully acknowledged. We also wish to thank Stephen Hall for useful comments and suggestions. The usual disclaimer applies. This paper aims to assess whether Europe is an optimum currency area (OCA) by examining synchronisation of business cycles and long-run output linkages in the EU countries. We
argue that a necessary condition for the desirability of EMU membership for national economies is that the degree of persistence of shocks affecting them should be similar. Given the low power of unit root tests, we measure instead the relative importance of permanent vs transitory components in output. The existence of a "European business cycle" is confirmed by correlation and cointegration analysis respectively. Finally, it appears that monetary coordination, by reducing exchange rate volatility, results in more synchronised cycles. EMU is therefore likely to be a successful experience, since the benefits of monetary integration will outweigh the costs of surrendering the exchange rate instrument. Keywords:
Optimum Currency Areas (OCAs), European and Monetary Union (EMU), Output Fluctuations, Permanent and Transitory Shocks, Correlation, Cointegration
JEL Classification:
E42, F36, F42
ISSN No 0969-6598
Contents
1.
Introduction ....................................................................................................1
2.
Optimum currency areas.................................................................................2
3. 3.1 3.2 3.3
Transitory vs permanent components in European outputs............................5 Models of output fluctuations.........................................................................5 Measuring the size of the random walk component in output........................6 A possible interpretation...............................................................................11
4. 4.1 4.2
Is there a European business cycle?..............................................................11 Output comovements....................................................................................11 Does synchronisation reflect the workings of the ERM? .............................13
5.
Conclusions ..................................................................................................14
References..............................................................................................................16
Tables.....................................................................................................................21
Charts .....................................................................................................................29
1. Introduction Much of the debate on the desirability of a monetary union in Europe has been based on the theory of "optimum currency areas" (OCAs), initially put forward by Mundell (1961) and McKinnon (1963). A prominent example is the report on "One market, one money" prepared by the European Commission (1990), which largely adopted this framework to analyse the economics of EMU and its potential costs and benefits. The basic idea is that optimality depends on whether or not countries are affected by symmetric real disturbances. If such shocks are dominant, then it is in each country's interest to forsake nominal exchange rate flexibility in order to benefit from the lower transaction and information costs that are associated with a single currency. Conversely, if asymmetric shocks prevail, the degree of labour mobility is low and nominal wages adjust slowly, EMU membership is not welfare increasing, as the cost of forgoing the exchange rate as an instrument to achieve relative prices adjustments outweighs the other benefits. In recent years a number of studies have attempted to give empirical content to the theory of OCAs, by analysing the various OCA criteria, namely the symmetry of shocks (see Bayoumi and Eichengreen (1996)), regional specialisation (see Krugman (1993)), labour market adjustments (see Eichengreen (1993)), fiscal federalism (see Sala-i-Martin and Sachs (1992)), and relative prices (see Eichengreen (1992)). In this paper we focus on the first criterion, and address two issues. Firstly, are output fluctuations in the EU countries idiosyncratic or synchronised? Secondly, if we observe business cycle conformity among EU countries' GNPs, is this due to the functioning of the ERM? Answering these questions requires investigating the sources of output fluctuations, and whether shocks hitting the European economies have permanent or transitory effects. We argue that a necessary condition for the desirability of EMU membership for national economies is that the degree of persistence of output shocks affecting them should be similar. If, instead, the stochastic properties of their GNPs are vastly different, then the optimality criterion for participation in the Union is not satisfied. Our approach is the following. Given the low power of unit root tests, we try to measure the relative importance of permanent versus transitory components in output. The autocorrelation function suggests that the former are rather small in EU countries, whilst they dominate in non-EU countries. Estimates from the Yule-Walker equations also indicate that the variance of the former is smaller in the European countries. This is confirmed by using the non-parametric measure of persistence developed by Cochrane (1988). A possible reason is that the monetary rule adopted in the European countries was of the money targeting type (see West (1988)). The existence of a "European business cycle" and of long-run linkages is supported by correlation and cointegration analysis respectively. The distinction between permanent and transitory components in output differentiates our paper from other empirical investigations, such as the study by Karras (1996), where a model in first differences is estimated in order to evaluate the relative size of country-specific and common shocks driving the growth rate of output. This type of analysis is valid only if the series are I(1), which cannot simply be assumed, and the reported results are therefore questionable. As Germany has generally been considered the "anchor" of the system (see, e.g., Fratianni and von Hagen (1990)), we test pairwise cointegration between each national economy's and
German output. From such tests one can also infer whether there are long-run relationships between outputs of the other member states: if the GNP of both country A and B, say, is cointegrated with German GNP, then they will also comove in the long run. Finally, it appears that there is a significant relationship between exchange rate volatility and the degree of output comovement. This is a likely consequence of the coordination of monetary and exchange rate policy resulting from ERM membership, which has resulted in more synchronised business cycles. The plan of the paper is as follows. Section 2 briefly discusses the concept of an OCA, and the empirical literature focusing on the symmetry of shocks. Section 3 provides estimates of the relative size of the permanent and transitory components in European outputs. Section 4 examines whether or not there exists a European business cycle, and tries to establish whether synchronisation in business cycles amongst European countries can be attributed, at least partially, to the workings of the ERM. Section 5 offers some concluding remarks. 2. Optimum Currency Areas The adoption of a single currency will have both benefits and costs. The former will be mainly in the form of lower transaction costs and of the disappearance of currency risks. The latter will be due to the inability of national governments and central banks to pursue independent monetary policies to stabilise the economy. The extent to which the loss of this policy instrument will affect the adjustment to equilibrium will depend on the degree of flexibility of factor markets and the nature of the shocks hitting the economy: the more rigid factor markets and the more country-specific the shocks, the more important will be the loss of monetary autonomy. These issues are addressed by the theory of optimum currency areas (OCAs), whose implications are crucial to answering the question whether Europe should adopt a single currency. If exchange rates are fixed, and factors of production are not sufficiently mobile, high adjustment costs, in terms of higher unemployment and lower output, are incurred in the presence of asymmetric shocks. The arguments for and against monetary integration are not as straightforward as those concerning economic integration more generally, as they hinge on issues such as credibility and coordination, transaction costs and bounded rationality (see Krugman (1989)). In its study "One market, one money", the European Commission (1990) relied heavily upon the theory of OCAs to evaluate the economic impact of EMU, although it also stressed the relevance of the literature on externalities and information costs (see Baldwin (1990)), and on policy games (see Barro and Gordon (1983), and Currie, Holtham and Hughes Hallett (1989)). The Commission concluded that EMU would ensure more microeconomic efficiency and macroeconomic stability. The improvement in the latter was put down especially to a reduction in the degree of exchange rate instability, and to higher credibility of the monetary authorities. The Commission, however, recognised that there would be disadvantages to EMU in terms of macroeconomic stability in the presence of asymmetric shocks. It also pointed out that, while country-specific shocks are by definition asymmetric, common shocks may affect national economies either symmetrically or asymmetrically, and that the main factors determining whether or not their impact is symmetric are the degree of product market integration, and differences in economic behaviour and structures. The empirical evidence
2
suggests that sectors producing homogeneous goods with few trade barriers mainly experience symmetric shocks. In the other sectors, there appears to be an inverse correlation between the existence of trade barriers and the degree of symmetry of the shocks. The seminal paper on OCAs is due to Mundell (1961). In his definition, an optimum currency area is an economic unit where factors of production are mobile and whose regions are affected symmetrically by shocks. Mundell stressed that the degree of labour mobility should be the main criterion to judge the optimality of forming a currency area (although workers might be reluctant to relocate across borders (see Ishiyama (1975)). High factor mobility (capital as well as labour) is not the only criterion proposed in the theoretical literature on currency unions to assess their desirability (see Masson and Taylor (1993)). For instance, the reduction in transaction costs is positively related to the degree of interdependence between economies in terms of trade flows - the higher this is, the greater will be the benefits from a monetary union. A related result is due to McKinnon (1963), who showed that in more open countries the exchange rate is less effective as a policy instrument, and hence its loss represents a lower cost. Krugman (1989) argues that the costs of fixing the exchange rates are outweighed by the benefits if there is intensive trade within the currency area. A further criterion is based on the degree of diversification of the economy (see Kenen (1969)). The more specialized countries are in the production of different goods, the more likely it is that shocks will be asymmetric, the more costly it is to forgo exchange rate flexibility. Finally, an issue not considered in the original literature is the flexibility of wages and prices. There is plenty of evidence that real wages are quite rigid in Europe, which implies that the real exchange rate is also quite rigid (see Eichengreen (1993)). Consequently fluctuations in the nominal exchange rate do not have large real effects in the labour market, and its irrevocable fixing (or, equivalently, the adoption of a single currency) would not be particularly costly in terms of unemployment1. Many empirical studies have tried to estimate the symmetry of disturbances across Europe2. The already cited report "One market, one money", Cohen and Wyplosz (1989), Weber (1990), and Garrett (1995) all examined the correlation of output movements. Using principal components analysis, Caporale (1993) showed that asymmetric shocks account for a sizeable percentage of GDP fluctuations in the EU. Karras (1996) also decomposed output fluctuations into common and country-specific shocks, and concluded that the latter are large and asymmetric. However, macroeconomic fluctuations reflect both shocks and policy responses. Hence a different approach is required to identify the former. Bayoumi and Eichengreen (1993a, 1993b, 1994, 1996) utilised the method developed by Blanchard and Quah (1989) to estimate a structural VAR, which is based on the assumption that demand and supply shocks are uncorrelated, and that only the latter have permanent effects on output. They found that supply shocks are larger in magnitude and less correlated across regions in Europe than in the US, and that there is a "core" of EU members (Germany, Austria, Switzerland, France, Denmark, Benelux) whose supply shocks are highly correlated, and 1
Note that this result only holds for the evaluation of real shocks, and it assumes that changes in the nominal rate do not have permanent real effects. If some assets held abroad are fixed in nominal foreign currency then a change in the nominal rate appears to have real effects. Many shocks are nominal, and removing the nominal exchange rate removes a nominal shock absorber. 2 For more extensive surveys, see Brociner and Levine (1992), Mason and Taylor (1993), Tavlas (1994), and Bayoumi and Eichengreen (1996).
3
another group of countries, the "EU periphery", with bigger and more country-specific shocks (see also Chamies et al (1994), and Erkel-Rousse and Melitz (1995)). Funke (1995) reported that even within the "core" correlations are lower if the sample period is extended to include German unification, and Bayoumi and Prasad (1995), and Funke (1997) found rather asymmetric region-specific demand and supply shocks across the EU countries. The international correlation structure of business cycles has also been examined by De Grauwe and Vanhaverbeke (1993), who found that asymmetric shocks tend to be more prevalent at the level of regions within a country than at the level of nations within Europe. Their analysis uses the standard deviation of the difference in percentage changes in income rather than the correlation of percentage changes in income. Artis and Zhang (1995) reported that most European countries' outputs were more highly correlated with the US during 1961-1979, and with Germany during the ERM period, the UK being the only exception. In particular, they used alternative detrending methods, i.e. the phase-averagetrend (PAT) methods and the Hodrick and Prescott (1980) filter (HP), and looked at synchronisation, phase shift and linkage between business cycles. Frankel and Rose (1996), again using a variety of detrending methods (such as fourth differencing, quadratic detrending, HP filtering), presented econometric estimates which suggest that there is a strong positive relationship between the degree of bilateral trade intensity and the cross-country bilateral correlation of business cycles. Other studies have looked at the variability of real exchange rates as an indication of the asymmetry of the shocks, as this should be related to demand or supply shifts across countries. The evidence is mixed (see, e.g., Poloz (1990), Eichengreen (1990), De Grauwe and Vanhaverbeke (1993)). Real share prices, which should reflect the present value of present plus expected future profits, have also been used, and a higher degree of divergence has been found in Europe than in other currency areas, implying that shocks are less symmetric across Europe (see Eichengreen (1992)). Finally, a number of papers have used disaggregated data by industry. Bini Smaghi and Vori (1992) come to the conclusion that in Europe shocks are predominantly industry-specific rather than country-specific, which makes EMU viable. By constrast, Funke, Hall and Ruhwedel (1997) report that, although on the whole country-specific shocks have become less important relative to worldwide shocks in the last twenty years, that is not equally true of all European countries. Therefore EMU should be restricted, at least initially, to those states with more economic homogeneity, as suggested, inter alia, by Dornbusch (1990) in his "two-track" EMU proposal. One should be aware that that the theory of OCAs is potentially subject to the Lucas critique. A regime shift, such as the creation of EMU, implies a change in the institutional structure of policy-making and the incentive structure of policy-makers, and, consequently, in the nature of shocks generated by policy actions. Therefore, the suitability of EMU for European countries cannot be evaluated using historical data: some countries might satisfy the necessary conditions for EMU entry ex post rather than ex ante, EMU membership being the catalyst of change. This argument has been put forward by Frankel and Rose (1996), who point out that, for instance, trade patterns and income correlations (two of the OCA criteria) are endogenous; furthermore, there is a significant positive relationship between trade integration and output comovements. However, this type of criticism is more likely to apply to temporary (demand) shocks, which are policy-induced, than to permanent (supply) shocks, which are associated with economic structure. The traditional theory of
4
OCAs has also been criticised by Bofinger (1994), on the grounds that past real exchange rate changes are not a good predictor of future real shocks, and issues such as credibility, the response to asymmetric shocks, and the effectiveness of monetary targets and instruments need to be addressed for an adequate evaluation of an OCA in Europe. 3. Transitory Vs Permanent Components In European Outputs 3.1 Models Of Output Fluctuations Traditionally, fluctuations in GNP were viewed as temporary deviations from a linear trend, which was interpreted as "potential GNP", determined in the steady state by capital stock, labour force and technology. Standard Keynesian models or monetary models with adaptive expectations and agents' misperceptions allow anticipated monetary policy to affect output, thus superimposing business cycles as short-run disequilibrium phenomena. Moreover, cyclical fluctuations in output can be produced by unanticipated monetary policy changes in stochastic equilibrium models which assume rational expectations as well as imperfect information (see Lucas (1972), and Barro (1976)). Therefore, detrending GNP data prior to analysing business cycle fluctuations became a routine practice. However, Nelson and Plosser (1982) and Campbell and Mankiw (1987a, 1987b) argued against the stationarity of GNP around a trend, claiming that output series contain a stochastic trend or a unit root. This meant that permanent supply shocks, rather than temporary demand shocks (fiscal or monetary), account for output movements. In other words, there is no meaningful distinction between the short run and the long run, and traditional Keynesian models should be abandoned in favour of Real Business Cycle (RBC) models, where the driving force is given by technology shocks (see, e.g., Long and Plosser (1983), and Prescott (1986), King et al (1988)). Empirical tests seem unable to reject the unit root null for a number of aggregate macroeconomic time series, including GNP, even though this cannot be easily justified in the context of representative agent models (see, e.g., Durlauf (1989), and Sims (1988)). Evaluating the adequacy of alternative theoretical models by examining the statistical properties of output is not without difficulties. It is now known that unit root tests, as those developed by Dickey and Fuller (1981), have low power. In any finite sample, it is impossible to discriminate between a root equal to unity and a root very close to this value (see Campbell and Perron (1991), and McCallum (1993)). In fact, as West (1988) notes, standard unit root tests, as those employed by Nelson and Plosser (1982), are not likely to reject the unit root null even when the true process is stationary with the sum of its autoregressive coefficient as low as 0.8. An alternative strategy is to decompose time series into a random walk and a stationary component, and to measure their relative importance (see Beveridge and Nelson (1981), and Cochrane (1988)). Even if it were possible to measure the degree of persistence of output an infinite degree of precision, one might still not be able to discriminate among competing economic theories. The reason is that a small random walk component is consistent with traditional business cycle models, but also with stochastic equilibrium models, which predict temporary fluctuations around a trend, once negative productivity shocks are assumed (see Kydland and Prescott (1982), and Long and Plosser (1983)). For instance, West (1988) shows that a large random walk component can be a property of models with purely nominal shocks.
5
Blanchard and Quah (1989) also analyse cases in which demand disturbances, such as changes in fiscal policy, have a long-run impact on output by affecting the savings rate, and subsequently the long-run capital stock. However, they argue that these effects are small compared to those of supply disturbances and suggest a methodology for identifying demand and supply shocks. It should be clear then that the presence of a unit root in GNP is compatible with models with totally different implications for stabilisation policies (see Durlauf (1989)). What is directly relevant for policy makers is the degree of persistence of (either demand or supply) shocks to GNP. In many models output persistence is generated by coordination failures (see, e.g., Diamond (1982)). If persistence is high, there is an argument for active countercyclical policy, since the cost of allowing low output levels would also be high. However, such policies might have persistent costs (such as inflation) as well as benefits (see Sims (1988)). 3.2 Measuring The Size Of The Random Walk Component In Output In the present study we are interested in establishing whether (i) the EU countries are characterised by a sufficient degree of business cycles conformity, and (ii) whether synchronisation is a product of the functioning of the ERM. A necessary condition for the former is that the stochastic properties of output be similar (i.e. that the size of the random walk component be comparable) across member states. Idiosyncratic business cycles will be observed, instead, if EU outputs exhibit different degrees of persistence, regardless of the sources of fluctuations. A sufficient condition for synchronisation in output movements can take two alternative forms, depending on whether the random walk component is small or big. In the first case, the stationary component in output accounts for most of the variance in output growth, and standard asymptotic theory, based on trend stationarity, may be a better approximation in a small sample. Therefore, one can detrend the output series, and then examine the size of the correlation coefficients between the detrended output series. The larger the correlation coefficient, the more suitable a country appears to be to join EMU. If the random walk component is big, then a modified asymptotic theory, based on unit roots, is required for valid inference. One should then examine the (pairwise) cointegration properties of the output series. What does cointegration imply for the suitability of individual countries to participate in a monetary union? Following Durlauf (1989), one could adopt a productivity interpretation of persistence, which stresses technology diffusion between countries. In this case, finding a big random walk component in all countries' outputs, and also cointegration among EU as well as non-EU countries, would have no implications for the creation of a monetary union. However, establishing cointegration only among the EU countries, combined with the finding of small random walk components, would be difficult to rationalise in the context of a productivity interpretation of persistence. Instead, it would be more plausible to argue that, in the presence of demand shocks of the type described above, EU membership has resulted in synchronised stabilisation policies among the participating countries, which explains why only EU (as opposed to non-EU) outputs are cointegrated. Let {yt} be a stochastic process defined on a probability space (S,F,P(.)). Let us also assume that it has the following autoregressive representation:
6
y t = a 0 + a1t + by t −1 + ut
ut ~ NIDO(0, σ 2 ), tε T
(1)
Depending on whether b is less than or equal to one, model (1) gives rise to trend stationarity or homogeneous nonstationarity (the unit root case). In particular, if a1=0 and b=1, then model (1) is a random walk with drift, and the unconditional mean and variance of {yt} are linear functions of time. If a1 ≠ 0 and b=1, then again a unit root exists, but now the unconditional mean of {yt} is a quadratic function of time. In both cases {yt} is neither an asymptotically independent nor an asymptotically stationary process. On the other hand, if b
