The New Economy and Global Stock Returns

Contributor Notes

Authors’ E-Mail Addresses: rbrooks2@imf.org, lcatao@imf.org

This paper revisits the relative importance of global versus country-specific factors underlying stock returns. It constructs a new firm level data set covering emerging and developed markets and estimates a simple factor model, which breaks down stock returns into a global business cycle factor, global industry factors, country-specific factors and firm-level effects. The results indicate that the share of variation in stock returns explained by global industry factors has grown sharply since the mid-1990s, at the expense of country-specific factors. Foremost among the global factors is a “new economy” factor, which has become a key determinant of global stock returns.

Abstract

This paper revisits the relative importance of global versus country-specific factors underlying stock returns. It constructs a new firm level data set covering emerging and developed markets and estimates a simple factor model, which breaks down stock returns into a global business cycle factor, global industry factors, country-specific factors and firm-level effects. The results indicate that the share of variation in stock returns explained by global industry factors has grown sharply since the mid-1990s, at the expense of country-specific factors. Foremost among the global factors is a “new economy” factor, which has become a key determinant of global stock returns.

I. Introduction

A longstanding empirical regularity in international equity markets is the low correlation among country portfolio returns. A number of explanations have been advanced to explain this stylized fact. First, instead of diversifying across markets and holding a portfolio that mirrors the global basket of securities, investors exhibit home bias in selecting stocks. If the marginal investor in French stocks lives in France and the marginal investor in U.S. stocks lives in the U.S., with each investor pricing stocks relative to other assets in the home market, country portfolios may in part reflect the different sentiment of French and U.S. investors. Second, country portfolios differ in industrial composition. For example, relative to Switzerland the Swedish stock market contains more firms in basic industries, while the Swiss index has more banks. To the extent that basic industries and banks are imperfectly correlated, the country indices of Sweden and Switzerland will be imperfectly correlated. Third, economic shocks may affect companies differently across countries. This may be because shocks are regional in nature, such as a policy change that is specific to one country. Alternatively national markets may respond differently to global shocks because institutional differences affect the transmission of these shocks to asset values.

Prior empirical work by Beckers et. al. (1992) and Heston and Rouwenhorst (1994) has shown that differences in the industrial makeup of countries play only a minor role in explaining the low degree of co-movement across national stock markets. Instead the low correlation of country portfolio returns is found to be primarily due to country-specific shocks. In other words, shocks that affect banks in Sweden differently than banks in Switzerland are more important in explaining the low correlation of national equity markets than the fact that Sweden has fewer banks. Or perhaps it is that cross-country variation in investor sentiment drives a wedge between the returns of companies that are in the same industry but in different countries. Heston and Rouwenhorst (1995) and Rouwenhorst (1999) show that these country-specific sources of return variation are dominant even in geographically concentrated and economically integrated regions such as Western Europe. They argue that country effects are likely to be even more important for stock markets that are further apart or in emerging markets, a hypothesis confirmed by Griffin and Karolyi (1998) and Serra (2000).

At the same time, there is a growing conviction in the investment community and in the financial press that globalization and the new economy are raising the importance of global industry effects in explaining return variation, at the expense of country-specific factors. A recent article in Business Week (09/11/2000) makes this point. It reports that the correlation between the S&P 500 and the Morgan Stanley Capital International Europe-Asia-and-Far-East (EAFE) index has increased from 25 percent in 1995 to 78 percent this year. The magazine interprets this increase in co-movement in the context of three phenomena: the wave of cross-border mergers and acquisitions, which is increasing the number of multinationals and accelerating the trend towards global industry sectors (consolidation within industries has accounted for three-quarters of all cross-border mergers over the past two years, against half in the early 1990s); the growing importance of high-tech companies, which are especially global in their reach (40 percent of Yahoo ! ‘s customers are outside the U.S., while Finland’s Nokia has a 37 percent share of the U.S. cellular market); and finally the fact that the internet makes it easier for investors to gather information on foreign companies, reducing the rationale for home bias in portfolio composition. Business Week concludes that, because companies are going global and national stock markets are increasingly correlated, diversifying across countries no longer offers investors the same amount of protection it once did. Perhaps as a result, the number of funds that invest globally by industry has been growing recently—14 of these funds were launched in just the past two years, with another 11 having filed registration statements to open soon.2 To quote one fund manager: “The world has changed, and the industry dimension matters more now than the country dimension.”

This perception on the part of the investment community has been partly corroborated by recent research. Baca et al. (2000) provide evidence that, while country factors remain more important, “pure” industry factors have grown in importance in the 1995-99 period across seven developed countries. Using a broader sample of developed countries, Cavaglia et al. (2000) find that the importance of industry factors has not only grown sharply in recent years but in fact outweighs that of country factors during the 1997-99 period. This paper extends these studies in two ways. First, it uses a new and much broader data set that covers up to 5,507 firms in 21 developed and 19 emerging markets, and accounts for around 90 percent of stock market capitalization across sample countries according to the 2000 IFC stock market fact book. We use this data to estimate a dummy-variable factor model of stock returns similar to that used in previous studies. Specifically, the model distinguishes between four kinds of factors: a global effect that captures broad co-movement across stock returns, in effect controlling for a global business cycle; country-specific effects that control for national determinants of stock returns; global industry-specific effects, which reflect the technological and product market characteristics of 10 broad sectors as defined by the FTSE; and global size effects, which control for risk-premia associated with smaller firms. Second, we use this model to measure the relative importance of a “new economy” factor in determining global stock returns. This allows us to examine the anecdotal evidence referred to above, that the new economy is promoting the importance of sectoral relative to country diversification in portfolio strategies.

The main results are as follows. First, the importance of global effects in explaining return variation has increased across the board since the mid-1990s. Meanwhile country-specific effects associated with developed countries have lost some explanatory power over the same period, while those for some emerging markets have increased dramatically in the wake of the Asian financial crisis of 1997-98. While the growing explanatory power of global factors may be seen as an indication that equity markets have become more integrated, it is also possible that the greater return variation explained by global factors is simply capturing that stock markets become more tightly correlated during crisis periods. Second, the fraction of return variation explained by global industry effects is on average 28 percent across stock markets in developed countries from mid-1997 onwards, far above the 7 percent identified by Heston and Rouwenhorst (1995). Even including emerging markets, the return variation explained by global industry factors amounts to 23 percent, far in excess of the 4 percent of Griffin and Karolyi (1998). This is clear evidence that industry sectors are becoming more important in diversifying portfolio risk. Third, a global industry factor associated with information technology far outpaces all other global factors in explaining return variation, a likely indication that it is not simply capturing tighter correlations due to the Asian crisis but the disparate behavior of technology stocks relative to the market as a whole. As a result, this paper finds evidence to support the notion that the new economy is raising the profile of industry sectors in portfolio diversification strategies. More broadly, this emerging “high-tech” effect suggests that the market bifurcation between old and new economy stocks—so apparent in some national equity markets—is in fact a global phenomenon that began as early as 1995. Fourth, the growing importance of the global information technology effect is robust across different specifications, notably for equal-versus value-weighting. It is also the case that the results are qualitatively unchanged when global size effects are added to the factor model, i.e., when we try to control for the possibility that an industry in one country may be different from the same industry in another country, using firm size to do this.

In a nutshell, the key result of this paper is to identify the growing importance of the global industry factor associated with the disparate behavior of technology stocks and their remarkable co-movement across markets. However, the paper is mute on whether this phenomenon reflects changing fundamentals or a global bubble. If this trend reflects changing fundamentals, the results suggest that the new economy revolution began earlier and is more global than previously thought. If the phenomenon is a bubble, it is possible that high-tech stocks are becoming a new conduit for financial contagion at the global level.

The remainder of the paper is structured as follows. Section II describes the data. Section III explains the factor model, while Section IV discusses the estimation results. Section V concludes.

II. The Data

The data cover monthly total returns and market capitalizations for up to 5597 firms in 21 developed and 19 emerging stock markets over the period March 1986 to August 2000.3 Firms are grouped into one of 10 FTSE industry sectors: resources, basic industries, general industries, cyclical consumer goods, non-cyclical consumer goods, cyclical services, non-cyclical services, utilities, information technology and financials. The appendix provides a detailed breakdown of the composition of these broad industry groups. While several recent papers argue in favor of a finer industry disaggregation, the level of disaggregation used here is sufficient because it follows to the traditional industry breakdown used by portfolio managers and contrasts information technology - a key “new economy” sector - with “old economy” sectors.4

Central to this paper is that the data be a realistic and unbiased representation of the global stock market. Table 1 provides a snapshot of the sample in December 1999.5 In that month the total number of firms amounts to 5,507, while the total number of listed firms in the 40 sample countries was 35,044 in that month, according to the 2000 IFC stock market fact book.6 The sample covers only 16 percent of listed firms. Looking at market capitalization measured in US dollars the picture is very different however. The total capitalization of the sample comes to $30,749 billion, almost 90 percent of stock market capitalization in the 40 sample countries as measured by the IFC. Coverage deteriorates somewhat towards the beginning of the sample. In December 1990, for instance, the total number of firms amounts to 3,177, while the total number of listed firms in the 40 sample countries was 22,535, again according to the IFC. The sample covers only 14 percent of listed firms. Total capitalization of the sample comes to $6,224 billion, about 67 percent of stock market capitalization in the 40 sample countries as measured by the IFC. In part, the deterioration in coverage highlights two important deficiencies of the data set. First, it is subject to survivorship bias, meaning that only firms surviving over the full sample period are covered. No doubt this bias is important, especially in the context of global shocks such as the Mexican and Asian crises. But it is most likely offset at least in part by the fact that the data omit a large number of small firms where the risk of bankruptcy is greater. A potentially more serious flaw of the data is that it includes only post-merger companies, dropping companies that go into the merger. In the case of the merger between Mercedes-Benz and Chrysler, for example, our data cover Daimler-Chrysler but not Mercedes-Benz and Chrysler individually in the period leading up to the merger. It is possible that this may bias the estimates in favor of finding more pronounced global industry effects in more recent years in the sample.

Table 1A.

Data on Number of Firms by Country and Industry, December 1999

article image
Table 1B.

Data on Market Capitalization in Billions of US Dollars by Country and Industry, December 1999

article image
Table 1C.

Data on Market Capitalization in Percentage of the Sample Total by Country and Industry, December 1999

article image

On the positive side, the sample is far more global and comprehensive than data used in earlier studies. No single country is represented by less than 31 firms (Peru) and, in the case of large economies such as the US and Japan, coverage approaches 1,000 firms towards the end of the sample. This large cross-section dimension of the data probably eliminates any significant distortion in the econometric results arising from the deficiencies mentioned above. Moreover, to the extent that much of our analysis focuses on the variance of stock returns rather than on their mean values, the main results are not particularly vulnerable to the traditional survivorship bias problem.

III. The Model

Following Heston and Rouwenhorst (1994) the model assumes that the return on each stock depends on four components: a global market factor (α), global industry factors (β), country factors (γ) and a firm-specific disturbance (e). The return on stock i that belongs to industry j and country k is given by:

Rit=αt+βji+γkt+eit(1)

The paper estimates a time-series for the realization of the common factor, industry factors and country factors by running the following cross-sectional regression every month:

Ri=α+j=1JβjIij+k=1KγkCik+ei(2)

where Iij is a dummy variable that equals one if the stock belongs to industry j and zero otherwise, and Cik is a similar dummy variable that identifies country affiliation. There are J industries and κ countries in total. This month-by-month approach is equivalent to a panel regression that interacts the constant as well as the industry and country effects with a time dummy—allowing for changes in the relative importance of these underlying factors over time. This is analogous to a seemingly unrelated regression model, which imposes no structure on the variance-covariance matrix that implicitly links the monthly regressions.

Equation (2) cannot be estimated in its present form because it is unidentified due to perfect multicollinearity. Intuitively, this is because every company belongs to both an industry and a country, so that industry and country effects can be measured only relative to a benchmark. To resolve this indeterminacy the paper follows the literature in imposing the restriction that the weighted sum of industry and country effects equal zero at every point in time, so that the industry and country effects are estimated as deviations from the intercept α.

j=1Jβji=iNIijxi=j=1Jβjwj=0(3)
k=1Kγki=1NIikxi=k=1Kγkvk=0(4)

where N is the total number of firms in a given month. Equation (2) is estimated using weighted least squares, with each stock return weighted by its beginning-of-month share of (sample) world stock market capitalization xi. Then wj corresponds to the market capitalization of industry j as a share of the (sample) global market, while vk is the market capitalization of country k as a share of the total.

The interpretation of these coefficients is straightforward. The intercept α reflects the return on the value-weighted portfolio of stocks across all sample countries – a benchmark against which industry- and country-specific effects are measured. Because (2) is estimated month-by-month, α will vary over time, capturing the impact of the global business cycle on stock returns across industries and countries. The estimated industry and country coefficients represent excess returns relative to this return. For example, βj measures the excess return on a portfolio of stocks in industry j, which is diversified to the same degree as the value-weighted global portfolio across countries. Similarly γk is the excess return on a portfolio of stocks in country k with the same industry composition as the value-weighted global portfolio. As long as no two countries in the sample have exactly the same proportion of firms across industries, there is no identification problem in estimating these industry-neutralized country effects and country-neutralized industry effects simultaneously.7

This model is useful in explaining differences in stock market performance across countries. The weighted sum of stock returns for country k has three components: the global factor, the weighted sum of industry effects and a country effect.

R^k=α^+j=1Jβ^jwjk+γ^k(5)

Equation (5) shows that there are two reasons that country performance differs from that of the global portfolio. The first is that industrial composition differs across countries. The weights wkj differ across countries (wkj is the share of industry j in the total market capitalization of country k), so that depending on industrial composition countries are subject to different industry effects. The second is the country effect, which accounts for differences in the return on stocks in country k relative to stocks in the same industry but located in another country.

A key deficiency of this approach is that it cannot accommodate interaction terms between industry and country effects. This is important because industry effects may be country-specific, especially if markets are segmented or during crisis episodes. We go beyond Heston and Rouwenhorst (1994) in trying to address this issue, using size dummies to proxy for country-specific industry effects, where the cross-section of $ capitalizations is divided into quintiles and affiliation denoted using dummy variables. As above, they weighted sum of coefficients is restricted to zero, so that the coefficients represent size risk-premia that are measured relative to the return on the (sample) world value-weighted portfolio. It is hoped that these size coefficients will control for the fact that a bank in Indonesia, say, is different from a bank in Germany, using its lower $ capitalization as a proxy.

The literature has used the factor model represented in equation (2) in three ways to measure the relative importance of country and industry effects in determining stock returns. Given that the explanatory variables are orthogonal by construction, one approach is to compare the R2 in (2) once one of the variables is omitted with that of the full model (see, e.g., Beckers and al. 1996). The difference in the cross-section of explanatory powers then measures the contribution of the omitted variable to explaining stock returns in a given period t. Breaking the sample into distinct sub-periods and averaging those cross-section R2s, it is possible to check whether the contribution of a given factor is rising or declining over time.

A second measure of the relative importance of industry versus country effects is to compare the average absolute value of the coefficients β^j and γ^k (Heston and Rouwenhorst, 1995, Rouwenhorst, 1999). If the mean of the absolute values of β^j across industries is smaller than that of γ^k across countries over a given period, this is indicative of a lower importance of industry- relative to country-effects during that period. Also, looking at those mean estimates across countries and industries over time sheds some light on what is driving the predominance of country or industry effects. For instance, a rise in the absolute mean value β^ for a specific sector k (e.g. information technology) may explain the growing importance of global industry relative to country factors.

Third, the relative importance of the distinct factors can be measured by the time-series volatility of the factor estimate. As the factor loadings in the model are either zero or unity, the explanatory power of a factor can be simply measured by the factor return variance. This permits testing several hypotheses regarding global market integration, the importance of industry versus country factors in the overall sample, as well as identifying which sector(s) or countries appear to explain the rising importance of one factor relative to another. For instance, if the global stock market is becoming more integrated over time, the variance of the global factors—the global market, industry, and size factors—should be increasing relative to the variance of the country factors.8 Using (5), we can measure, on a country by country basis, the proportion of the total variance in stock market returns R^k that is explained by changes in the global return α^ versus changes in industry composition effects (j=1Jβ^jwjk) or changes in the country factor γ^k. And, again, splitting the sample and computing such variances over sub-periods allows us to test whether such global industry effects have in fact grown in importance, the extent to which this phenomenon has been more pronounced in certain countries (or group of countries), and whether such a growing importance of global factors has been mostly due to the particular behavior of information technology or any other individual sector(s).

Each of these measures provides useful information that is in some respects complementary. For instance, the cross sectional R2 statistic gives us the net percentage contribution of country, industry, and size factors to explaining stock returns at a given point in time, since industry, country and size dummies are orthogonal in every cross-section. However, the same statistic does not allows us to measure how much of the time series variations in returns is explained by changes in the global factor α^ (as the latter is fixed every cross-section), and is therefore insufficient to gauge the degree of global market integration. Measuring the latter requires estimates of mean excess returns for each industry/country (i.e. the second measure discussed above) and their respective Sharpe ratios, as will be argued below. Finally, measures of volatility decomposition based on equation (5), while allowing us to measure the contribution of global and industry composition factors to overall stock returns in each country, can only provide an estimate of their gross contributions since the right-hand side variables in (5) are not orthogonal over time.9 In light of the limitations of these different measures as well as their complementarities, we shall look at all of them, rather than singling out one as some previous studies have done.

IV. Results

As noted above, both the equal- and value-weighted regressions use monthly total returns expressed in U.S. dollars.10 The value-weighted regressions weight returns by beginning-of-month U.S. dollar capitalization as a share of the (sample) global stock market.

Table 2 reports the average R2 for equal- and value-weighted models for different combinations of explanatory variables. Those R2s are cummulative, i.e., the first row in each sub-section of the table presents the R2 for equation (2) setting the dummy I to zero, whereas the two subsequent rows report the R2 of the models including I and C, and then including I, C and the size dummy. Thus, the net contribution of the additional factor (industry or size) to the cross-section of stock returns will be given by the difference between the R2 reported in two successive rows. In presenting these results we breakdown the sample into two – one covering developed countries only and the other including emerging markets as well. The rationale for focusing only on developed economies in the first set of regressions is twofold. One is data heterogeneity. Data for several emerging markets only becomes available in the 1990s, whereas data for mature economies spans the entire sample period. The other reason is that most previous studies on the relative importance of country versus industry effects have focused on mature markets. Thus our results for mature markets allow a more direct comparison with the preceeding literature.

Table 2:

Average R2 for Each Model

article image

For the equal weighted model, the results of the R2 statistic are very similar to those of Becker et al. (1996) in that industry factors account for only 3 to 4 percent of the cross sectional variation of stock returns in developėd countries. However, the same equal weighted model points to a declining importance of country factors in explaining stock returns throughout the 1990s (both among developed countries and for the whole sample), while the importance of the industry-affiliation factor rises (albeit marginally) during the 1990s.

A better fit but different estimates of the contribution of each factor are obtained on the basis of value-weighted regressions. Intuitively, these differences are in fact to be expected since the equal-weighted model exaggerates the importance of emerging markets and of lower capitalization firms in the composition of the global portfolio. To the extent that the dispersion of returns in those countries and firms tend to be more heterogenous cross-sectionally, the role of both the global market factor α and global industry factor β tend to be downplayed. Accordingly, table 2 shows that industry factors explain a much higher proportion of the cross-sectional dispersion of stock return variations in the value-weighted model, both for the whole sample as well as among industrial countries. Moreover, the value-weighted model points to a more significant rise in the net explanatory power of industry effects over the 1990s. A counterpart of it is the decline in country effects which, again, is steeper than in the equal-weighted regressions, though also in the same direction. As will be seen below, this steep decline in country effects is consistent with the findings of the variance decomposition analysis based on equation (5) which points to the greater role of the global market factor α^ and of the industry-specific coefficients β’s in explaining the total time-series variance of stock returns in recent years. In all regressions and sub-periods, size effects turned out to be far less important than country- or industry-affiliation effects.

Tables 3 and 4 present period averages of the cross-section mean estimates and the time-series variance of those mean estimates by industry and country. Table 3 covers developed countries only while table 4 reports results for the whole sample including emerging markets.11 Each table gives the means for the global, country, and industry factors over the relevant period, the standard deviations of each of these factors over time, and the corresponding Sharpe ratios. As explained above, while the means of the industry and country factors represent excess returns or risk-premia relative to the global (sample) portfolio, the importance of each factor is given not only by the absolute value of its mean, but also by the standard deviation of this excess return over time, as the latter reveals how much variation in stock returns is explained by a particular factor.

Table 3:

Value-Weighted $ Returns Model: FTSE Industry Effects-Industrialized Countries

(US Dollar returns in percentage per month)

article image
Table 4.

Value-Weighted Percentage Returns Model: FTSE Industry Effects

(US dollar returns in percentage per month)

article image
article image