«ABSTRACT Over the period 1972-1986, the correlations of GDP, employment and investment between the United States and an aggregate of Europe, Canada ...»
Financial Globalization and Real Regionalization
Stern School of Business, New York University, NBER and CEPR
Over the period 1972-1986, the correlations of GDP, employment and investment between the
United States and an aggregate of Europe, Canada and Japan were respectively 0.76, 0.66,
and 0.63. For the period 1986 to 2000 the same correlations were much lower: 0.26, 0.03, and
-0.07 (real regionalization). At the same time, U.S. international asset trade has signiﬁcantly increased. For example, between 1972 and 1999, United States gross FDI and equity assets in the same group of countries rose from 4 to 23 percent of the U.S. capital stock (ﬁnancial globalization). We document that the correlation of real shocks between the U.S. and the rest of the world has declined. We then present a model in which international ﬁnancial market integration occurs endogenously in response to less correlated shocks. Financial integration further reduces the international correlations in GDP and factor supplies. We ﬁnd that both less correlated shocks and endogenous ﬁnancial market development are needed to account for all the changes in the international business cycle.
keywords: International business cycles, international diversiﬁcation jel classication codes : F36, F41 We thank Ariel Burstein, Marco Del Negro, Robert Engle, Paul Evans, Alessandra Fogli, Philippe Martin, Andy Neumeyer, two anonimous referees and seminar participants at the CEPR ESSIM 2001, the 2001 ES Summer Meetings, the 2001 Duke International Economics Group Meetings in Venice, the 2001 SIEPR Meetings in Stanford, the 2002 AEA meetings, Carnegie Mellon, MIT, New York University, Ohio State University, Queens University, UC Santa Barbara, Rutgers, the University of Washington, and Wharton for helpful comments. Remaining errors are our own.
1. Introduction Over the past 30 years, the United States economy has increasingly danced to its own tune.
Over the period 1972 to 1986 the business cycle frequency correlations of output, employment and investment between the U.S. and an aggregate of Europe, Canada and Japan were 0.76, 0.66, and
0.63 respectively. For the period 1986 to 2000, the corresponding correlations were 0.26, 0.03, and
-0.07. The consumption correlation also declined between the two periods, but to a smaller extent (from 0.51 to 0.13). We call this phenomenon real regionalization. At the same time, trade in international ﬁnancial assets has sharply increased. Between 1972 and 1999, United States gross holdings of foreign direct in investment and equity in the same group of countries rose from 4 to 23 percent of the U.S. capital stock. We label this trend ﬁnancial globalization.
In this paper we argue that changes in the international business cycle and growth in international asset trade are intimately related. In particular, increasing globalization in ﬁnancial markets is key to accounting for less international co-movement. We present evidence that the correlation of the stochastic shocks hitting the U.S. and the rest of the world has fallen in the post-Bretton Woods era. We then consider model economies in which the degree of international diversiﬁcation is endogenous, and show that a fall in the correlation of shocks increases equilibrium diversiﬁcation by increasing the potential gains from international asset trade. Finally, we investigate whether a combination of less correlated shocks coupled with the resulting deepening of international asset markets can account for the observed changes in the international real business cycle.
The ﬁrst model we consider is a simple atemporal two-country endowment economy. At the start of the period agents trade shares in domestic and foreign assets which deliver imperfectly correlated dividends. A shipping cost associated with foreign dividend income constitutes an incentive to bias portfolios towards the domestic security. Reducing the correlation of dividends across countries leads to greater diversiﬁcation, an increase in the cross-country correlation of consumption relative to the correlation for output, and an increase in the volatility of net exports.
We then proceed to consider a richer inﬁnite-horizon model with capital in order to assess the extent to which a calibrated model economy can capture both the quantitative extent of observed growth in international asset trade, and at the same time account for changes in cross-country correlations of the same magnitude as those observed empirically. The assets that may be traded internationally are shares in a representative domestic ﬁrm and a representative foreign ﬁrm. We consider two calibrations of the model corresponding to the early high-shock-correlation period, and the more recent period in which shocks have been less correlated.
In response to the fall in the shock correlation we ﬁnd that the equilibrium level of portfolio diversiﬁcation increases, and that this increase is of the same magnitude as our empirical measures of the change in ﬁnancial integration. Reducing the correlation of the shocks without allowing agents to adjust their portfolios has the eﬀect of reducing the international correlations of macro-aggregates, but not by as much as these correlations fell empirically. However, the endogenous increase in international portfolio diversiﬁcation that arises in equilibrium further reduces the international correlations of output, employment and investment. Thus we ﬁnd that a combination of the fall in the correlation of productivity shocks and the resulting endogenous growth in international asset trade can jointly account for most of the observed changes in the international business cycle.
Our work is closely related to the work of several authors including Baxter and Crucini (1995), Arvanitis and Mikkola (1996), Kehoe and Perri (2002) and Heathcote and Perri (2002) who discuss the implications of limited international trade in ﬁnancial assets in international real business cycle economies. None of these papers, however, addresses the eﬀects of growth in foreign asset holdings on business cycle dynamics.2 There are various papers that analyze the role of international diversiﬁcation for other issues. For example, Obstfeld (1994) describes a model in which an increase in diversiﬁcation increases the economy’s growth rate by encouraging a switch into high-risk high-return investments. In Martin and Rey (2001) the set of assets traded is endogenous, and economic integration has implications for risk sharing and asset prices.
There are few papers that document changes in international business cycle regularities over time. One of them is Kollmann (2001) which examines earlier changes in international business cycle correlations. Finally, a long run perspective on ﬁnancial globalization is oﬀered by Obstfeld and Taylor (2002).
2. Data In this section we present various measures of the international correlation of business cycles in the post-Bretton Woods period. We then report statistics on international diversiﬁcation and the volume of international trade in ﬁnancial assets.
A. United States and world business cycles One measure of the change in the international business cycle correlation in the post-Bretton Woods period (1972 - 2000) is the diﬀerence between cross-regional correlations in two equal length subsamples. Table 1 displays the correlations of business cycle frequency ﬂuctuations in GDP, consumption, investment and employment between the U.S. and an aggregate of the rest of the world (comprising Europe, Japan and Canada).3 For details on the data see the data appendix. Correlations are computed on the log of Hodrick-Prescott ﬁltered quarterly series. The numbers in parentheses are the standard errors when estimating the correlation coeﬃcients using GMM (see Backus and Kehoe 1992).
Notice ﬁrst that the correlations of all variables have markedly declined. The correlations of investment and employment have fallen the most, while consumption is the variable which exhibits the smallest drop in correlation. While a general decline in correlations might be simply due to a decline in the correlation of exogenous shocks (the 1970’s were dominated by world-wide oil shocks), the relatively large falls in the correlations of investment and employment suggest a change in the asset market structure. In particular, development of international ﬁnancial markets increases opportunities for intertemporal specialization in production, and thus might be an important factor in accounting for the reduced correlation in factor supplies.
We ﬁrst document that the decline in correlation is robust to alternative detrending procedures, to changes in the sample period, and to possible bias in measures of correlation due to heteroskedasticity. We then document in more detail the change in correlation across diﬀerent regions.
Table 2 shows that the drop in international correlations is robust to measuring business cycle ﬂuctuations by ﬁrst diﬀerencing logged data (taking growth rates) or by taking deviations from a linear trend. Both alternative detrending procedures show a large drop in the correlation of inputs of production, and a relatively modest fall in the consumption correlation.
Table 2. Alternative detrending procedures
The ﬁrst two rows of table 3 shows that the overall reduction in correlation is still present when considering a longer time period starting in the ﬁrst quarter4 of 1962. In this case the ﬁrst subsample includes the 1960’s, which were a decade of weaker international correlations.5 The second two rows report correlations for a shorter time period starting in 1980.1. In this case the ﬁrst sub-sample excludes the common oil-shock dominated 1970’s, but the decline in cross-country correlations remains.
62.1-81.2 0.53 0.30 0.60 0.57 1962-2000 81.3-00.4 0.34 0.15 -0.02 0.14 80.1-90.2 0.58 0.25 0.22 0.54 1980-2000 90.3-00.4 -0.16 -0.10 -0.41 -0.14 Additional evidence that our ﬁnding does not depend on the particular time period chosen is presented in ﬁgure 1. The four lines in the ﬁgure are the rolling correlation estimates for conditional correlations of output, consumption, investment and employment.6 The ﬁgure reveals that internaThe ﬁrst quarter of 1962 is the ﬁrst date for which we have complete series for all countries from the OECD.
The fact that in the 1960’s the international correlation of business cycles was quite low has been noticed by various authors. Kollmann (2001) has used this fact to assess the importance of monetary shocks in accounting for business cycles.
See, for example, Engle (2000). In quarter t the estimate of the conditional correlation is the correlation between the two series over the interval t − n to t. We set n = 58, which is the same sample length used in tables 1 and 2.
tional correlations for all variables were relatively stable (and high) until the mid 1980’s, since when they have declined steadily.
Several authors have recently pointed out that in sub-samples with high conditional volatility, estimates of the correlation between variables tend to increase, even though there are no changes in the underlying distribution (see, for example, English and Loretan 1999). If we measure the change in business cycle volatilities in the U.S. and in our aggregate corresponding to the rest of the world (table 4) we ﬁnd a marked reduction in volatility, especially for the U.S.
We therefore use the procedure proposed by Forbes and Rigobon (2002) to correct the measure of business cycle correlation using estimates of standard deviations in the two sub-samples.7 Corrected correlation estimates are presented in table 5. Notice that although the drop in correlation is reduced, it is still signiﬁcant (in particular for investment and employment), suggesting that
where ρ is the standard correlation coeﬃcient between x and y in the subsample, and ∆ is the ratio of the variance of x in the subsample to the variance of x in the whole sample. The formula above is an exact estimate of ρ∗ only in the case of joint normality of x and y. In general our data do not satisfy this property, but Forbes and Rigobon (2001) argue that even in more general cases the correction above gives a relatively good estimate of the true correlation.
Table 5. International correlations (corrected measure)
Finally, to assess how robust the decline in cross-country correlations is to alternative country groupings, we report correlations between U.S. macro variables and their counterparts in Europe, Japan and Canada. Table 6 indicates that comovement between the U.S. on the one hand and either Europe or Japan on the other has declined signiﬁcantly. However, the U.S. and Canadian business cycles are now more correlated than in the ﬁrst half of the sample.
Table 6. International correlations
B. International trade in ﬁnancial assets In this subsection we document the increase of U.S. trade in international assets. The measure of international diversiﬁcation we focus on for U.S. assets is the sum of the U.S. foreign direct investment position (at current cost) plus the equity part of the stock of portfolio investment abroad, relative to the U.S. capital stock (see the data appendix for more details). The analogous measure of liabilities is the ratio of the sum of the stock of foreign direct investment in the U.S. plus foreigners holdings of U.S. stocks to the U.S. capital stock. Consistently with the business cycle evidence presented in the previous section, we focus on U.S. holdings of assets in Western Europe plus Canada and Japan, and these countries’ holdings of U.S. assets. The measure of the capital stock we use is the net stock at current cost of private non-residential assets.