Journal Issue
Share
Article

Currency Unions, Economic Fluctuations, and Adjustment: Some New Empirical Evidence

Author(s):
International Monetary Fund. Research Dept.
Published Date:
January 1997
Share
  • ShareShare
Show Summary Details

The issue of whether Europe constitutes an optimum currency area increases in significance as the Maastricht process for moving to a single currency progresses. There are many ways of approaching this question, one of which has been to contrast behavior across U.S. regions with that across European countries.1 By comparing behavior within the United States with that within Europe, researchers have attempted to provide information on how the behavior of European economies corresponds to that of an existing and successful currency union of similar economic size. If European countries do relinquish their own currencies, they will face many of the same constraints faced by U.S. regions. Hence, such a comparison could also shed light on the potential effects of a new currency in Europe.

This paper compares economic fluctuations in Europe and the United States, using a more comprehensive data set than in previous literature. Parallel data sets comprising output, employment, and productivity differentiated at the broadly defined one-digit sectoral level are constructed for U.S. regions and for eight European countries since the early 1970s. These data are used to look at two criteria for an optimum currency area. The first is the level of industrial diversification and the relative importance of region-and industry-specific disturbances in economic fluctuations. Exchange rate changes can mitigate the effects of disturbances that affect all industries in a given region. By contrast, the value of the exchange rate as a method of adjustment is diminished if disturbances are primarily industry specific, particularly if the industrial structure is relatively diversified. Investigating sectoral diversification and the nature of underlying shocks is therefore an important ingredient in assessing the suitability of a single currency.

The second issue that we investigate is the level of labor market integration in the United States and Europe. Adjustment to underlying disturbances is as important an issue with respect to optimum currency areas as the nature of the disturbances themselves. Indeed, labor market mobility was the criterion that Mundell (1961) focused upon in his seminal contribution to the literature on optimum currency areas.

The first section provides further motivation by reviewing recent work on regional adjustment, particularly in the United States. Section II describes the data and presents some results on industrial diversification in the United States and Europe. Section III discusses the econometric methodology. The results from our analysis of disturbances are reported in Section IV, while Section V presents the results on labor market adjustment. Section VI concludes.

I. Regional Adjustment

Recent work on regional adjustment in the United States pertaining to Economic and Monetary Union (EMU) contains a number of strands. Blanchard and Katz (1992) examine U.S. state-level data on employment, wages, and unemployment and conclude that employment bears the brunt of regional adjustment in the United States.2 A negative disturbance that lowers employment in a given state produces relatively little real wage response. Rather, the labor market regains equilibrium as the excess labor moves to a new location within the United States. The implication is that in the United States interregional labor mobility is the major equilibrating force in the economy. In Europe, by contrast, the main equilibrating mechanism over the short run appears to be changes in the labor force participation rate (Decressin and Fatas, 1995).

Regional diversification of industries has also been examined. By comparing industrial diversification in the United States with that in the European Union (EU), possible effects of EMU on European economic geography can be inferred. Krugman (1991) concludes that the greater regional specialization exhibited by industries in the U.S. manufacturing sector relative to those in European countries is a function of the common currency and, hence, that over time EMU may imply significant regional dislocation.3 A different perspective is provided by the Commission of the European Communities (1990), which argues that increasing integration of the EU will make western Europe a better candidate for a currency union.

Finally, there have been a number of comparisons of the behavior of underlying disturbances that drive economic fluctuations in the EU and the United States, but little consensus on the lessons to be learned from such a comparison. For example, Bayoumi and Eichengreen (1993) examine data on aggregate output by region and country and conclude that disturbances within the EU as a whole are less correlated than those within the United States, implying significant costs to monetary union. Bini Smaghi and Vori (1992) use data on output across 11 manufacturing industries and find that industry-specific shocks account for the majority of the explained variance in output in both the United States and Europe. As the exchange rate is not a potent instrument for dealing with industry-specific disturbances, they conclude that the exchange rate is not a particularly useful adjustment mechanism in Europe.4

This paper is concerned with industrial diversification and the nature of, and adjustment to, underlying disturbances in the United States and the EU. Before discussing our approach, a number of limitations in any comparison of the United States with the EU as a guide to the impact of EMU should be recognized. The institutional structures in the two regions are different. The United States has a much more important federal fiscal system, a single language, a unified cultural heritage, lower taxes, fewer state enterprises, and a weaker tradition of government intervention in the economy than most EU countries. In addition, the United States has operated with a common currency for over 200 years, so that the analysis will have little to say about the speed or difficulty of the economic transition implied by moving from separate monies to a currency union. Finally, the level of regional inequalities within the United States is somewhat lower than across EU countries.

At the same time, the similarities of the underlying economic structures in the United States and the EU (outside the monetary field) should also be recognized. Both are continent-wide economies, with similar levels of development, population, and per capita income. Both are characterized by mature, market-based economies and democratic political institutions. When aggregated into a single economy, the EU is, like the United States, relatively closed to international trade.5 Hence, while not being the only factor at work, it is probably not unreasonable to attribute a significant portion of the observed differences in behavior to the existence of a unified currency in the United States and separate national currencies in the EU.

II. Industrial Diversification

Parallel data sets were constructed for the eight standard U.S. regions defined by the Bureau of Economic Analysis (BEA) and for seven EU countries plus Austria, which, although not in the EU during our sample period, has close economic ties to Germany and has recently joined the Union. The data set consists of three variables—real output (value added), employment, and output per employee—and covers eight industrial classifications: primary industries (or mining, where data on agriculture were not available); construction; manufacturing; transportation; trade; finance; services: and government. The U.S. data come from the BEA regional data bank. The European data come from the National Accounts of the Organization for Economic Cooperation and Development (OECD), and the real output data were converted into U.S. dollars using 1985 purchasing power parities, also obtained from the OECD.6 The data are annual and generally cover the period 1970–89 for the United States and 1970–87 for the EU. However, some of the employment (and, hence, productivity) series were available only for a slightly shorter time period.

Table 1.Comparison of Industrial Classifications
ClassificationU.S. regionsEuropean countries
PrimaryAgriculture, forestry, and fisheries plus miningAgriculture, hunting, forestry, and fishing plus mining and quarrying
ConstructionConstructionConstruction
ManufacturingManufacturingManufacturing
TransportationTransportation and public utilitiesTransport, storage, and communication plus electricity, gas, and water
TradeWholesale trade plus retail tradeWholesale and retail trade, restaurants, and hotels
FinanceFinance, insurance, and real estateFinance, insurance, real estate, and business services
ServicesServicesCommunity, social, and personal services
GovernmentGovernmentGovernment services

The United States and the OECD use somewhat different industrial classification systems, and it was necessary to amalgamate some series to produce industrial sectors that were more closely aligned. Table 1 shows the aggregation that was used, based on the major industrial classifications in each data set. Although some differences in classification still remain,7 the result is a pair of data sets whose classifications are. we believe, compatible enough to be used for comparative work.

Table 2 reports some comparative statistics across the two data sets. It shows the average share of total output produced by each industry within the region or country, as well as the mean and the coefficient of variation of these industry output shares. The mean values illustrate the composition of output across different industries. Many industries have similar mean ratios across the two data sets. However, the services sector is significantly more important in the United States than in Europe, while manufacturing is more important in Europe. Manufacturing has the largest share of output in both data sets.

Table 2.Industry Output Shares: U.S. Regions and EU Countries
PRMCTNMFRTSPTRDFINSVCGVTTotal
U.S. region
New England.01.05.26.08.16.17.17.11.06
Mideast.01.05.21.10.16.17.17.13.20
Great Lakes.03.04.31.09.16.14.13.09.18
Plains.10.05.21.10.17.15.13.10.07
Southeast.09.06.21.09.16.13.12.13.20
Southwest.22.06.15.09.14.12.11.10.11
Rocky Mountains.15.07.13.11.15.14.13.13.03
Far West.05.05.19.08.17.16.17.13.15
Mean.08.05.21.09.16.15.14.12
Coefficient of variation.87.15.28.11.06.12.17.14
EU country
Austria.04.09.29.09.17.15.04.15.04
Belgium.03.07.23.11.19.07.16.14.05
Denmark.08.07.19.09.16.17.05.20.03
Germany.03.07.35.08.11.12.12.12.36
Greece.16.05.20.13.13.08.13.12.03
Italy.05.08.24.11.19.12.09.13.24
Netherlands.13.06.20.08.13.16.11.13.08
United Kingdom.07.06.26.10.15.19.05.15.18
Mean.06.07.28.09.14.14.09.13
Coefficient of variation.67.19.22.17.21.32.48.20
Notes: The totals in the final column indicate the average share of each region (or country) in total U.S. (or EU) output. The means and the coefficients of variation for industry output shares are reported in the final two rows of each group. For each group, industry output is divided into eight classifications: primary industries (PRM): construction (CTN): manufacturing (MFR); transportation (TSP); trade (TRD); finance (FIN): services (SVC): and government (GVT).
Notes: The totals in the final column indicate the average share of each region (or country) in total U.S. (or EU) output. The means and the coefficients of variation for industry output shares are reported in the final two rows of each group. For each group, industry output is divided into eight classifications: primary industries (PRM): construction (CTN): manufacturing (MFR); transportation (TSP); trade (TRD); finance (FIN): services (SVC): and government (GVT).

The coefficient of variation is a measure of the degree of regional specialization of an industry. The larger the variation in the composition of output across regions, the larger the coefficient of variation. Primary industries and manufacturing in the United States are highly concentrated in particular regions, presumably due to the concentration of agriculture and mining in the Plains, Southwest, and Rocky Mountain regions and of manufacturing in the Great Lakes region. The European countries in our sample show less specialization in these two industries. In all other industries, however, the coefficient of variation is higher in the EU than in the United States.8

Based on an examination of manufacturing sector data, Krugman (1991) concludes that the United States is a significantly more specialized economy than Europe. He therefore argues that the introduction of a single currency in Europe would create an impetus toward greater specialization and. consequently, lead to significant reallocation of labor and other factors following EMU. The results in Table 2 do not support this argument. Our measure of specialization indicates that, if anything, EU countries are somewhat more specialized than U.S. regions in industries other than manufacturing and primary goods, at least at the one-digit Standard International Trade Classification (SITC) level. Manufacturing may, therefore, not necessarily provide an adequate basis for comparing the structure of the U.S. and EU economies.9 Outside of manufacturing, the pertinent concern for EMU may not be Krugman’s argument that greater specialization will create changes in industrial structure. To the contrary, the greater homogeneity in industrial structure engendered by EMU could well be a more potent factor.

More generally, the lack of any clear pattern in the results in Table 2 in terms of the relative specialization of Europe and the United States implies that industry-specific disturbances are likely to have broadly similar effects in these two economic areas. Hence, if there is a significantly different economic impact from underlying disturbances, it must come from the nature of the disturbances themselves. It is to this topic that we now turn.

III. Econometric Methodology

This section presents the econometric methodology that we employ to identify the sources of disturbances: those that affect all industries within a given region or country (regional shocks); those that affect industries across all regions or countries (industrial shocks); and those that affect all regions or countries and all industries simultaneously (aggregate shocks). Such a decomposition allows us to analyze the nature of the disturbances affecting the United States and the EU, and how these two economic areas adjust to these disturbances.

Our data sets contain observations on output, employment, and productivity over time, disaggregated by U.S. region or EU country and by one-digit industry. The combination of eight industries and eight regions or countries in each data set implies a panel with a maximum of 64 observations per time period, each identified by industry, location, and date. The sources of the underlying disturbances are measured, using the following specification:

where Δln(yi, j, t) is the change in the logarithm of output in industry i, region/country j, and period t; αi, t, βj, t and ψt are the coefficients associated with dummy variables that are equal to 1 for industry i in period t, for region/country j in period t, and for all industries and regions in period t, respectively (and zero otherwise); and εi, j, t is an error term.10

Researchers primarily interested in the propagation of business cycles have often used richer dynamic formulations as they wish to identify innovations to output (or employment) growth.11 By contrast, our primary interest is in identifying the sources of observed output fluctuations rather than their propagation, so as to assess the suitability of a single currency for Europe. Hence, we do not include additional dynamic terms in our specification, although we do later test for dynamic interactions across the identified disturbances.

In the specification above, if the αi, t coefficients were calculated for all industries i, a linear combination of these coefficients would be equal to the time-specific dummy variable ψt. The same is true of the region/country dummies βj, t, if summed over all j. Accordingly, one industry and one region need to be eliminated from the set of dummy variables to identify the model. The choice of the omitted industry and region/country does not affect tests of the significance and explanatory power of the industrial or regional effects. An F-test of the joint significance of the remaining αi, t coefficients represents a valid test of the importance of industry-specific shocks in the regression, as does a similar test of the joint significance of the βj, t coefficients.

Since the industry-specific and region-specific dummy variables are orthogonal by construction, the explanatory power of these variables can also be calculated from the reduction in the R2 statistic caused by excluding them from the original regression. Any variation that is explained by the regression but that is not specifically attributable to either set of dummy variables can be attributed to the aggregate disturbance.12

The exclusion of one of each of the αi, t and βj, t coefficients is of more importance when the estimated coefficients are used to construct a series that represents the underlying disturbances of industry i or region/country j. As estimated, the series αi (made up of αi,1, αi,2, . . . ,αi, T) represents the shock to industry i relative to the shock to the industry that was excluded from the estimation. Similarly, the series βj represents the shock to region/country j relative to the excluded region/country, while the series ψ represents the sum of the aggregate disturbance plus the shocks to the excluded industry and excluded region/country. To distinguish the aggregate disturbance from that experienced by the excluded industry and region, a further restriction is necessary. The restriction employed here is that the sum of all of the αi, t disturbances (including the region excluded from the estimation) is equal to zero in each period t a similar restriction was imposed on the sum of the βj, t disturbances. The rationale is that the industrial and regional shocks represent deviations from an underlying aggregate disturbance and the aggregate impact of these deviations should then sum to zero. The aggregate disturbance itself was then calculated as the value ψ minus the implied shocks to the industry and region excluded from the estimated set of dummy variables.

In addition to decomposing short-term sources of fluctuations in output, we also consider the nature of labor market adjustment to these disturbances. Average rates of growth of output, employment, and output per worker over several years are used to calculate the relative importance of regional and industrial factors in labor market adjustment, using the following cross-sectional regression:

where Δln(y*i, j) is the average change in output for industry i in region j. Since there is no time dimension here, it is not possible to identify an aggregate disturbance. The analysis is therefore limited to the relative importance of regional and industrial factors in medium-term adjustment. Largely industry-specific productivity trends indicate a relatively high level of labor market integration, as such integration is needed to reduce productivity differentials across regions. Decomposing long-run employment growth into region- and industry-specific factors helps us to identify whether labor or capital is the main channel of factor mobility.

IV. Sources of Disturbances

The U.S. data cover the period 1972–89, implying that there are 1,152 observations (8 industries times 8 regions times 18 data points). The European data cover the period 1971–87 and contain 1,088 observations.

U.S. Results

Table 3 reports the overall explanatory power of equation (1) and the importance of industry-specific, region-specific, and aggregate disturbances in this total.13 As shown in panel C, equation (1) explains 73 percent of the variation in disaggregated U.S. output growth. The aggregate disturbance is the most important factor, explaining 29 percentage points of the variance, while the industrial and regional dummy variables explain a further 25 and 19 percentage points of the variance, respectively. F-tests of the significance of the dummy variables (not reported) indicate that all of the elements of the model (the industry, region, and time dummies) are highly significant. In short, the model as a whole explains three-fourths of the variance of output growth, and all three types of disturbances are significant, with the aggregate disturbance explaining the largest fraction, industrial factors being almost as important, and regional factors accounting for a smaller share.

Panel A of Table 3 also reports the overall R2 and the decomposition between the different factors for each industry; these results are calculated using the estimated coefficients from the full regression but limiting each calculation to only those observations that involve that industry. Regional disturbances turn out be relatively unimportant for the manufacturing industry; indeed, using this approach, the impact on the R2 is negligible.14

Table 3.Decomposition of Short-Term Fluctuations a(R2 attributable to various shocks)
U.S. regionsEuropean countries
TotalAggregateIndustryRegionTotalAggregateIndustryCountry
A. Industries
Primary.43.39.17Primary.26.31.08
Construction.80.36.31.34Construction.51.17.19.16
Manufacturing.83.67.24Manufacturing.71.38.15.19
Transportation.81.45.31.06Transportation.69.28.13.28
Trade.94.41.37.16Trade.62.28.07.28
Finance.61.13.16.33Finance.56.34.07.15
Services.85.49.11.26Services.27.20.09
Government.54.00.33.28Government.55.21.32.02
B. Regions/Countries
New England.70.30.08.31Austria.53.20.23.10
Mideast.75.26.22.27Belgium.54.19.19.16
Great Lakes.77.39.30.08Denmark.50.11.27.12
Plains.69.29.33.08Germany.62.33.20.09
Southeast.80.43.32.04Greece.57.27.11.19
Southwest.68.16.18.34Italy.45.22.33.10
Rocky Mountains.72.15.26.31Netherlands.49.21.22.07
Far West.73.33.32.08United Kingdom.48.06.41
C. Full Period and Subperiods (Aggregate Results)
Full period.73.29.25.19.52.19.18.16
1972–79.72.31.26.15.53.18.14.21
1980–87.76.29.26.21.51.14.24.13

Estimating equation: Δln(yi,j,t)=αi,t+βj,t+Ψt+εi,j,t.

Estimating equation: Δln(yi,j,t)=αi,t+βj,t+Ψt+εi,j,t.

These results suggest that sectoral factors are more important than regional factors in explaining variation in the growth of output in manufacturing. Thus, our aggregate results for U.S. manufacturing are consistent with the more disaggregated results, using two-digit industry classifications, obtained by Bini Smaghi and Vori (1992). A similar result is obtained for the transportation sector. By contrast, regional disturbances explain a significant part of the variance in construction, finance, services, and government, four industries that make up almost half of total output in the United States. Finally, the trade and primary sectors are an intermediate case, with results between these two extremes. These differences appear to be intuitive. Manufacturing and transportation, which produce goods that are easily traded across regions, are dominated by nonregionally differentiated shocks. Industries whose products are less easily traded geographically, such as construction, finance, services, and government, are more prone to regional disturbances.

The same decomposition can be carried out across regions. The results in panel B of Table 3 indicate that the relative importance of the three types of disturbances also vary by region. Regional disturbances are most important in the Southwest and Rocky Mountains regions, presumably because of the importance of raw material production in the local economies. The Mideast and New England, which are relatively specialized in finance and other service industries, also have relatively large regional disturbances. By contrast, regional disturbances are the least important factors in the Great Lakes and Plains regions, which are among the most specialized in manufacturing.

As discussed earlier, it is also possible (by putting the relevant αi, j and βi, j coefficients into a time series) to derive individual series for the underlying disturbances to the eight regions, the eight industries, and the aggregate disturbance. The top panel of Figure 1 plots the growth in total output and the aggregate disturbance for the United States. The two series are clearly highly correlated, indicating that our methodology has yielded a reasonable measure of the aggregate shock. The aggregate disturbance is negative after the oil price hikes in the 1970s and positive through much of the late 1980s. Visual inspection indicated that the other disturbances also appear sensible. For example, the disturbance for primary industries showed a positive impact from the oil price hikes and a negative pattern in the late 1980s, while the disturbance for New England vividly illustrated the rise and fall of the “Massachusetts miracle.” In this regard, the results for the EU also appear reasonable, as shown in the bottom panel of Figure 1, which plots the aggregate disturbance and total output growth for this area.

Panel A of Table 4 shows the correlation between the aggregate disturbance and the disturbances for individual industries, with statistically significant correlations marked with an asterisk. The disturbances for manufacturing and finance are significantly positively correlated with the aggregate disturbance, indicating that the cyclical effects of aggregate shocks are amplified in these two industries. By contrast, the disturbances associated with services and primary goods are negatively correlated with the aggregate. In the case of services, this negative correlation presumably reflects the dampening of aggregate fluctuations by this industry. For primary industries, it appears more likely that the negative correlation illustrates the opposite impact of commodity price changes (particularly in oil prices) on the fortunes of the industry and of the economy as a whole. Interindustry correlations (not reported) reinforce these results. In particular, disturbances between manufacturing and both services and primary goods are highly negatively correlated.

Figure 1.Aggregate Shock and Total GDP Growth a

a To make the series compatible, the annual rate of growth of total output was divided by 10 and its mean was subtracted.

Table 4.Correlations with Aggregate Disturbance
A. Industries
U.S. RegionsEU Countries
Primary-0.62*Primary-0.66*
Construction0.20Construction-0.07
Manufacturing0.73*Manufacturing0.56*
Transportation0.18Transportation0.30
Trade-0.38Trade0.39
Finance0.59*Finance0.65*
Services-0.59*Services-0.49*
Government0.27Government0.04
B. Regions/Countries
U.S. RegionsEU Countries
New England0.02Austria0.05
Mideast-0.05Belgium-0.02
Great Lakes0.35Denmark-0.27
Plains0.00Germany0.53*
Southeast0.63*Greece0.20
Southwest-0.21Italy0.12
Rocky Mountains-0.23Netherlands0.09
Far West0.12United Kingdom-0.34
Notes: An asterisk (*) indicates that the correlation coefficient is significant at the 5 percent level. Under the null hypothesis that the true correlation coefficient is zero, the approximate standard error of these coefficients is 0.24.
Notes: An asterisk (*) indicates that the correlation coefficient is significant at the 5 percent level. Under the null hypothesis that the true correlation coefficient is zero, the approximate standard error of these coefficients is 0.24.

The correlation coefficients between the regional disturbances and the aggregate (panel B of Table 4) are generally smaller than those associated with industrial disturbances, and the only significant correlation is the positive one between the Southeast and aggregate disturbances. Interregional correlations (not reported) indicate that New England and the Mideast face very similar disturbances, as do the Southwest and Rocky Mountains, but that disturbances between these two pairs of regions are large and negative. The U.S. economy appears to be divided into three distinct regions: the Northeast, the raw-material-producing central states, and the remainder.15

As noted earlier, we are more interested in identifying types of disturbances rather than the mechanisms through which these disturbances are propagated. However, we did examine the dynamic properties of the estimated disturbances. In general, the shocks did not display significant persistence overtime, with most of the first-order autocorrelation coefficients being small and insignificant. We also used bivariate Granger causality tests (with two lags) to examine if there were important feedback effects among the various disturbances. We found that the null hypothesis of no Granger causality could be rejected at the 5 percent significance level in only 7 percent (18/256) of the cases, suggesting that our methodology adequately captures the important dynamic properties of the data.

EU Results

A similar decomposition of output growth fluctuations was carried out for the eight European economies. Panel C of Table 3 shows that equation (1) explains about one-half of the total variation in the growth of disaggregated output in the EU, with the aggregate, industry-specific, and country-specific disturbances accounting for 19 percent, 18 percent, and 16 percent, respectively. Comparing the results for the United States and the EU, the relative importance of the different disturbances is strikingly similar. In both cases, industry-specific shocks contribute about one-third of the explained variance in output growth, with aggregate shocks contributing slightly more and country/region shocks slightly less. At the same time, it should be noted that the relative contribution of country-specific shocks is slightly larger in Europe than in the United States (31 percent of the explained variance in the EU versus 26 percent in the United States).

There are a number of differences from the U.S. results at the industry level in the EU countries. Country-specific factors account for more than one-fourth of the 71 percent of variance explained for manufacturing.16 Country-specific factors are also more important than industry-specific factors in transportation and trade, possibly because these industries are more concentrated spatially in the EU than in the United States (see Table 2). In construction and services, two nontraded goods sectors, industry-specific factors have more explanatory power than country-specific factors in the EU, the reverse of the result for the United States.

When the decomposition is carried out for each country (panel B of Table 3), large variations are seen in the relative importance of the three types of disturbances. Aggregate factors are most important in Germany, Greece, and Italy. Industry-specific factors are more important than country-specific factors in most countries, with the exception of Greece and the United Kingdom. In the United Kingdom, country-specific factors appear to dominate output growth fluctuations, suggesting that aggregate factors that affect other European countries have little impact.

Panel A of Table 4 reports the correlations between the aggregate disturbance and the disturbances for individual industries in the EU. As in the United States, the disturbances for manufacturing and finance amplify the aggregate shock, while the disturbances for the primary and service sectors are significantly negatively correlated with the aggregate disturbance. The correlation coefficients between the country-specific disturbances and the aggregate disturbance (panel B of Table 4) are generally not statistically significant. The notable exception is Germany, which has a strong positive correlation with the aggregate, presumably reflecting the importance of the German economy in the EU. There are few significant correlations among the interindustry disturbances (not reported here).

As in the case of the disturbances for the United States, the disturbances for the EU did not generally reveal significant persistence. Furthermore, bivariate Granger causality tests again confirmed the absence of important dynamic effects across disturbances.

We also examined the possibility of structural changes leading to variation over time in the relative importance of disturbances. For both data sets, we ran the regressions separately over identical subperiods, 1972–79 and 1980–87. The aggregate results are reported in panel C of Table 3 and show that the results are quite similar over the two subperiods. One interesting feature of these results is that, for the EU, industry-specific shocks are more important and country-specific shocks are less important in the 1980s than in the 1970s. The industry results (not reported here) reveal a similar picture. This outcome is consistent with the notion that trade flows and financial market integration have led to a greater degree of integration among European economies in the 1980s, thereby enhancing the suitability of a single currency for Europe (Commission of the European Communities, 1990).

In summary, the results for the United States and the EU reveal a similar aggregate picture of the relative importance of various sources of disturbances. However, a disaggregated perspective reveals an interesting difference. In the United States, region-specific disturbances are most prevalent in nontraded goods sectors, such as services and construction. By contrast, country-specific disturbances in the EU are important in traded goods sectors, although in all sectors the relative importance of country-specific disturbances has declined in the 1980s.17

V. Labor Market Adjustment

Thus far, we have analyzed the nature of disturbances to disaggregated output growth. An equally important issue is how economies respond to such disturbances. In particular, we focus on the degree of integration and nature of adjustment of labor markets in the United States and the EU by considering the determinants of long-term trends in output, employment, and productivity. These trends are decomposed into sectoral and regional components. If labor markets are highly integrated across regions, implying an absence of wage differentials, the levels of productivity should be independent of regional effects (assuming, as seems reasonable, that the same technology is used in a given industry across all regions). Hence, if trends in productivity primarily reflect the fortunes of particular industries, this would imply more integrated labor markets. By contrast, if such underlying productivity trends are primarily regional, this would imply a low level of labor market integration.18

The relative importance of regional and industrial disturbances in employment trends, on the other hand, indicates the degree to which labor markets equilibrate through firms moving to regions of excess labor supply (region-specific effects) or labor moving to expanding industries (industry-specific effects). Hence, productivity regressions measure the integration of labor markets, while employment regressions measure how the labor market adjustment that does occur is achieved.

Table 5.Long-Term Adjustment: 1972–89a(R2 attributable to different factors)
U.S. regionsEuropean countries
TotalIndustryRegionTotalIndustryCountry
Output.82.66.16.77.15.61
Output per worker.89.85.01.83.19.64
Level of output per worker.97.94.02.75.50.25
Employment.89.63.24.69.61.08
Notes: For the European countries, the sample ends in 1987. All dependent variables are used in growth rates in the regressions except for the productivity level regressions (level of output per worker).

Estimating equation: Δln(zi,j)=αi+βj+εi,j.

Notes: For the European countries, the sample ends in 1987. All dependent variables are used in growth rates in the regressions except for the productivity level regressions (level of output per worker).

Estimating equation: Δln(zi,j)=αi+βj+εi,j.

The underlying econometric approach is similar to that used to examine disturbances, except that the time dimension is excluded. The sample averages for each of the relevant variables (level of productivity and rates of growth of output, employment, and productivity) were calculated for each region and sector.19 For each of these variables, equation (2) was then estimated over the full sample (1972–89 for the United States and 1971–87 for the EU) and then over two subsamples: the 1970s and the 1980s.

Table 5 reports the results from the full sample. In the United States, the full regression explains over 80 percent of the variation in average rates of output growth over the 1972–89 period. Four-fifths of the explanatory power comes from the industrial dummies and one-fifth from the regional dummies. The performance of an industry within a region appears much more closely related to the overall performance of that industry rather than to the performance of that region. In short, industrial structure can go a long way toward explaining relative performance across regions in the United States.

The results for both levels and changes in productivity indicate that the contribution of the regional dummies to the overall regression is very small and, hence, that U.S. labor markets are highly integrated, at least over long time spans. Of the 97 percent of variation explained by the regression for productivity levels, the industrial dummy variables account for 94 percentage points, and regional dummy variables a mere 2 percentage points, while the remaining 1 percentage point is unallocatable.20 Despite the low level of explanatory power, an F-test indicates that the regional dummies are jointly significant at conventional significance levels.

The regressions for productivity growth show a similar pattern. Of the total explanatory power of 89 percentage points, the contributions of the industrial and regional dummy variables are 85 percentage points and 1 percentage point, respectively. Unlike the productivity levels regressions, however, the regional dummy variables are not jointly significant in this case.

The regressions for employment growth indicate a larger, although still subsidiary, role for regional factors. Slightly over one-fourth of the total explanatory power in the regression comes from the regional dummy variables, with the remainder attributable to their sectoral counterparts. The implication is that the majority of economic adjustment occurs through movements of labor to regions with expanding industries, rather than through movements of expanding industries to regions with excess labor. In short, regional labor market migrations of the type emphasized by Blanchard and Katz (1992) appear to be the predominant form of regional adjustment in the United States.

The results from the regressions for output and productivity growth for the EU are strikingly different. Although the regression for average output growth has about the same explanatory power as in the case of the United States, the relative contribution of country-specific factors is about four times that of the industry dummies, the reverse of the result for the United States. In the EU, the correlation of average output growth is much higher across industries within a given country than across countries for a particular industry.

The productivity regressions suggest that labor markets are far less integrated in the EU than in the United States. The regression using productivity levels shows that country-specific factors have a far more important role in this regression than in its U.S. counterpart. In the regression using the growth in productivity, more than three-fourths of the total explanatory power of the regression is attributable to country-specific dummies. Unlike in the United States, long-term trends in productivity in the EU appear to be overwhelmingly determined by national performance, rather than by industrial factors.

Table 6.Long-Term Adjustment: Subperiods a(R2 attributable to different factors)
U.S. regionsEuropean countries
TotalIndustryRegionTotalIndustryCountry
A. 1972–79
Output.80.43.37.59.07.52
Output per worker.90.88.01.75.20.55
Level of output per worker.95.93.03.78.38.40
Employment.85.23.52.50.44.06
B. 1980–89
Output.62.34.28.69.26.43
Output per worker.82.74.05.70.13.58
Level of output per worker.98.96.02.17.62.10
Employment.79.70.09.74.64.11
Notes: For the European countries, the sample ends in 1987. All dependent variables are used in growth rates in the regressions except for the productivity level regressions (level of output per worker).

Estimating equation: Δln(zi,j)=αi+βj+εi,j.

Notes: For the European countries, the sample ends in 1987. All dependent variables are used in growth rates in the regressions except for the productivity level regressions (level of output per worker).

Estimating equation: Δln(zi,j)=αi+βj+εi,j.

The employment regression shows that country-specific factors play a very small role in explaining differences in long-term employment growth. As in the case of the United States, this implies that long-term trends in employment are primarily determined by industrial factors. However, as the productivity growth regressions indicate that labor markets in the EU are not highly integrated across national borders, the interpretation of these results is different. Unlike in the United States, intersectoral reallocation of labor appears to operate only within, not across, EU countries.

Finally, we examine whether the patterns that exist over the full 1972–89 period can also be identified over somewhat shorter periods by repeating the analysis for two subperiods, 1972–79 and 1980–89.21 The results are reported in Table 6. In the United States, the regressions over shorter time periods confirm the lack of importance of regional factors in explaining either levels or changes in productivity. However, regional factors are generally more important in explaining changes in output and employment over these subperiods than over the full time period, probably because of the slow pace of labor market adjustment. If labor market adjustment is a gradual process, the importance of regional factors will decline over time. For the EU, the subsample results were very similar to the full sample results, suggesting that, from 1970 to 1987, there were no significant structural changes that affected the degree of integration of labor markets.

VI. Conclusions

This paper has analyzed the effects of a currency union on the relative importance of different types of shocks to output growth and also the labor market mechanisms by which economies adjust to these shocks. We constructed two comparable data sets for U.S. regions and eight European countries, with data on output, employment, and productivity at the one-digit sectoral level. Although the two data sets are similar in many respects, an important difference is the fact that the U.S. regions are part of a currency union while the European countries are not.

For the full sample, the relative importance of aggregate, industry-specific, and country- or region-specific shocks in explaining output growth fluctuations is roughly similar in Europe and the United States, as each of these types of shocks plays an important role.22 A more disaggregated analysis of the sources of disturbances at the sectoral level, however, indicates that region-specific disturbances in the United States are more important in non-traded goods sectors, while country-specific disturbances in the EU are more prevalent in traded goods sectors. In addition, the relative importance of country-specific disturbances has declined in me EU in the 1980s, plausibly reflecting moves toward economic integration over this period.

The major difference between the United States and the EU, however, is in the nature of labor market adjustment to shocks. Our results indicate that productivity trends are dominated by industry-specific factors in the United States and by country-specific factors in the EU. These results appear to confirm other evidence that the United States has a much more integrated labor market, either because of, or reflecting, the single currency.

Our regressions for long-term employment growth in the United States produced results consistent with the findings of earlier authors that interregional flows of labor constitute a more important adjustment mechanism in the U.S. labor market than labor flows across countries in Europe. This implies that large wage differentials across European countries could remain after EMU. In addition, unless labor mobility across European countries is enhanced, wage differentials across countries will have to remain flexible if significant disruptions from country-specific disturbances are to be avoided in EMU.

References

    AltonjiJosephG. andJohnC. Ham1990 “Variation in Employment Growth in Canada: The Role of External, National, Regional, and Industrial Factors,”Journal of Labor EconomicsVol. 8 (January) S198-S235.

    • Crossref
    • Search Google Scholar
    • Export Citation

    AtkesonAndrew andTamimBayoumi1993“Do Private Capital Markets Insure Regional Risk? Evidence from the United States and Europe,”Open Economies ReviewVol. 4 (July) pp.303-24.

    • Crossref
    • Search Google Scholar
    • Export Citation

    BayoumiTamim andBarryJ. Eichengreen1993 “Shocking Aspects of European Monetary Unification,”in Adjustment and Growth in the European Monetary Unioned. byFrancescoGiavazzi andFranciscoTorres (Cambridge: Cambridge University Press) pp.193-229.

    • Crossref
    • Search Google Scholar
    • Export Citation

    BayoumiTamim andBarryJ. Eichengreen1996“Operationalizing the Theory of Optimum Currency Areas” paper presented at Centre for Economic Policy and Research Conference on Regional Integration La Coruna SpainApril.

    • Search Google Scholar
    • Export Citation

    BayoumiTamim andPaulR. Masson1995“Fiscal Flows in the United States and Canada: Lessons for Monetary Union in Europe,”European Economic ReviewVol. 39 (February) pp.253-74.

    • Crossref
    • Search Google Scholar
    • Export Citation

    BayoumiTamim andGabrielSterne1993“Regional Trading Blocs, Mobile Capital and Exchange Rate Coordination,”Bank of England Working Paper Series No. 12 (London: Bank of England).

    • Search Google Scholar
    • Export Citation

    BeanCharles1992“Economic and Monetary Union in Europe,”Journal of Economic PerspectivesVol. 6 (Fall) pp.31-52.

    BiniSmaghiLorenzo andSilviaVori1992“Rating the EC as an Optimal Currency Area: Is It Worse than the U.S.?” in Finance and the International Economy,Vol. 6AMEX Bank Review Prize Essaysed. byRichardO. O’Brien (Oxford and New York: Oxford University Press for the AMEX Bank Review) pp.78-104.

    • Search Google Scholar
    • Export Citation

    BlanchardOlivierJ. andLawrenceF.Katz1992“Regional Evolutions,”Brookings Papers on Economic ActivityNo. I pp.1-75.

    Commission of the European Communities1990“One Market, One Money: An Evaluation of the Potential Benefits and Costs of Forming an Economic and Monetary Union,”European EconomyVol. 44 (October) pp.3-347.

    • Search Google Scholar
    • Export Citation

    DecressinJorg andAntonioFatas1995“Regional Labor Market Dynamics in Europe,”European Economic ReviewVol. 39 (December) pp.1627-55.

    • Crossref
    • Search Google Scholar
    • Export Citation

    EichengreenBarry1990“One Money for Europe? Lessons from the U.S. Currency Union,”Economic PolicyVol. 5 (April) pp.117-87.

    EichengreenBarry1992“Should the Maastricht Treaty Be Saved?” Princeton Studies in International Finance Paper No. 74 (Princeton, New Jersey: International Finance Section. Princeton University).

    • Search Google Scholar
    • Export Citation

    KrugmanPaulR.1991Geography and Trade (Leuven, Belgium and Cambridge, Massachusetts: Leuven University Press and MIT Press).

    LumsdaineRobin andEswarPrasad1997 “Identifying the Common Component in International Economic Fluctuations,”NBER Working Paper No. 5984 (Cambridge, Massachusetts: National Bureau of Economic Research).

    • Search Google Scholar
    • Export Citation

    MelitzJacques1996a“The Theory of Optimal Currency Areas, Trade Adjustment, and Trade,”Open Economies ReviewVol. 7 (April) pp.99-116.

    MelitzJacques1996b“The Evidence on the Costs and Benefits of EMU” (unpublished: Paris: Institut National de la Statistique et des Études Économiques).

    • Search Google Scholar
    • Export Citation

    MundellRobertA.1961“A Theory of Optimum Currency Areas,”American Economic ReviewVol. 51 (September) pp.657-65.

    NorrbinStefanC. andDonE. Schlagenhauf1988“An Inquiry into the Sources of Macroeconomic Fluctuations,”Journal of Monetary EconomicsVol. 22 (August) pp.43-70.

    • Crossref
    • Search Google Scholar
    • Export Citation

    NorrbinStefanC. andDonE. Schlagenhauf1996“The Role of International Factors in the Business Cycle: A Multi-country Study,”Journal of International EconomicsVol. 40 (February) pp.85-104.

    • Crossref
    • Search Google Scholar
    • Export Citation

    Sala-i-MartinXavier andJeffreySachs1992 “Fiscal Federalism and Optimum Currency Areas: Evidence for Europe from the United States,”in Establishing a Central Bank: Issues in Europe and Lessons from the USed. byMatthewB. CanzoneriVittorioGrilli andPaulMasson (Cambridge: Cambridge University Press) pp.195-219.

    • Search Google Scholar
    • Export Citation

    StockmanAlanC.1988“Sectoral and National Aggregate Disturbances to Industrial Output in Seven European Countries,”Journal of Monetary EconomicsVol. 21 (March-May) pp.387-409.

    • Crossref
    • Search Google Scholar
    • Export Citation

    vonHagenJürgen1992“Fiscal Arrangements in a Monetary Union: Evidence from the U.S.,”in Fiscal Policy Taxation and the Financial System in an Increasingly Integrated Europeed. byDonaldE. Fair andChristiande Boissieu (Dordrecht, Netherlands and Boston: Kluwer Academic Publishers) pp.337-59.

    • Search Google Scholar
    • Export Citation
2An earlier study by Eichengreen (1990), which looks at the behavior of unemployment across U.S. regions, comes to similar conclusions.
3A number of authors have also investigated mechanisms that cushion the effects of economic disturbances within the United States, including federal fiscal policy (Sala-i-Martin and Sachs, 1992; von Hagen, 1992; and Bayoumi and Masson, 1995) and private capital markets (Atkeson and Bayoumi, 1993).
4This argument has been repeated by a number of other authors. See, for example, Melitz(1996b).
5This is particularly true if the members of the European Free Trade Association are included in the European aggregate (Bayoumi and Sternr, l993).
6State-level data were aggregated into the eight standard BEA regions in order to make the U.S. data more comparable to the EU data in terms of the number of regions and their economic size. The eight U.S. regions are New England, Mideast, Great Lakes, Plains, Southeast, Southwest, Rocky Mountains, and Far West. The eight European countries are Austria, Belgium, Denmark, Germany, Greece, Italy, Netherlands, and the United Kingdom. Constraints on data availability led us to exclude other important European countries, such as France.
7For instance, hotels are classified in the service sector in the regional U.S. data and in the trade sector in the European data.
8The results in the text are based on data for the full sample period. To examine whether factors such as increasing European integration could affect our conclusions, we also constructed our measures of specialization for the first and last five years of the sample for both data sets. There were no important differences relative to the results for the full sample.
9Some of the apparent specialization within the EU may reflect problems in making industrial classifications consistent across countries. However, it is more likely to result from the wide diversity of regulations and practices across EU countries, which could mean that similar tasks are often carried out by different industrial sectors.
10Δln(yi, j, t) is measured as the deviation from the mean growth rate of the series as a whole in industry i, country/region j, at time t, thereby controlling for individual fixed effects. The specification assumes that region- or country-specific disturbances have the same effect on the growth rate of output in all industries. To control for differences in cyclical sensitivities across industries within each region/country, output growth rates for each industry i in each region/country j were divided by the sample standard deviation of output growth for that series. The decomposition is similar to that used by Stockman (1988), except that we include time-specific dummies ψt. Hence, in our setup, αi, t and βj, t can be directly interpreted as the orthogonal components of the industry-specific and region- or country-specific shocks, respectively.
11Norrbin and Schlagenhauf (1988 and 1996) and Allonji and Ham (1990).
12These dummy variables are exactly orthogonal only when all data points are available. In other cases, there is a small residual value that is unattributable across the three types of disturbances.
13As discussed earlier, to avoid collinearity, dummies for one industry (government) and one region (Far West region) are excluded from the U.S. regressions. Similarly, dummies for one industry (services) and one country (Italy) are excluded from the EU regressions. All of our results are quite robust with respect to the choice of excluded industry and country/region.
14Since we are looking at a subset of the original data set, it is possible for the independent variables to lower the variance of the dependent variable (of course, this is not possible for the full data set). In rare instances, this resulted in a small negative contribution for a factor. We set these R2 contributions to zero.
15Bayoumi and Eichengreen (1993) also find that the raw-material-producing regions face distinctly different underlying disturbances. However, they do not rind the same dichotomy between the Northeast and the rest of the economy.
16The relative importance of country-specific factors in manufacturing in European countries is similar to the findings of Stockman (1988), who uses two-digit manufacturing data. However, the results are different from those obtained by Bini Smaghi and Vori (1992), who conclude that sectoral factors account for a substantial fraction of variation in this sector’s output growth in the EU. Using different techniques and higher frequency data, Lumsdaine and Prasad (1997) find that about one-fourth of the total variation in monthly growth rates of industrial production in European OECD countries can be attributed to a common European component.
17Although an analysis of the causes of this pattern of shocks is beyond the scope of this paper, we did run some simple regressions of the disturbances on various measures of real exchange rate changes. Our preliminary results indicate no clear evidence of any relation between exchange rate changes and the estimated disturbances.
18Further evidence on the behavior of European labor markets can be found in Decressin and Fatas (1995).
19Levels of productivity were measured as the average of the logarithm of output per worker.
20Because there are some missing values, the two sets of dummy variables are not exactly orthogonal. Hence, some of the variance can be explained by either. The reported values are the marginal contributions of each set of dummy variables to the overall explanatory power, measured as the increase in the R2 that occurs when these variables are included in a regression already containing the other explanatory variables.
21Other shorter subperiods, not reported, showed broadly similar results. For the EU, data availability limited the analysis in the second subperiod to 1980–87.
22The importance of country-specific disturbances implies that the exchange rate could continue to be a potentially important tool in mitigating the effects of such disturbances.

Other Resources Citing This Publication