Against the backdrop of the euro and the trend toward globalisation in general, there seems to be a growing consensus among institutional investors that country diversification has become less relevant and sector diversification more important. I decided to investigate the proposition that sectors have grown in importance relative to countries.
This analysis is somewhat data-challenged, but I believe my data is sufficient to generate some preliminary findings. My data set consists of the monthly price returns from January 1987–March 2000 of six economic sectors in five countries, a total of 30 return series. The six sectors and five countries are listed in Table 1.
My choice of these sectors and countries was determined solely by the availability of data. I created five country return series by computing the average return of the six sectors within each country (for example, the average return of banks, chemicals, communications, electronics, retail, and transport in Canada). Then I created six sector return series by computing the average return of the five countries within each sector (for example, the average return of Canadian, German, Japanese, UK and US banks). You might argue that I should have weighted the components of these series by their relative capitalisations, but I did not have this information historically. The important point as I see it, however, is that in aggregate my collection of country returns comprises the same underlying securities as my collection of sector returns. They are simply stratified differently.
My first pass at the data compares the range of returns across countries and across sectors. I calculate a two-year average monthly return for each country and take the difference between the maximum and minimum returns. I then repeat this calculation for the sector returns. If countries were more important than sectors, we should expect to see a wider range of returns for countries than for sectors.
My next experiment compares the correlation structure of countries and sectors. I calculate the average correlation among country returns and among sector returns for the full sample of returns and for three non-overlapping sub-samples: January 1987–December 1991, January 1992–December 1996, and January 1997–March 2000. I also calculate rolling two-year correlations for both countries and sectors. In this experiment, we should expect to see a lower correlation for the more efficient method of stratification.
Finally, I compute the optimal portfolios associated with country and sector selection. I use the same return samples as I did for the correlation experiment, and in all optimisations I assume that risk aversion equals two. This assumption means that an investor is willing to concede two units of expected return in order to lower portfolio variance by one unit. In this experiment, the more efficient stratification method would be expected to produce a higher ratio of expected return to risk.
Figure 1 plots the range of average monthly returns across countries and across sectors for rolling two-year periods, and Figure 2 shows the difference in these two series (sectors minus countries).
Figure 1 shows that the range of country returns has been variable but relatively trendless and that the most recent range is close to the historic average. By contrast, the range of sector returns has remained fairly stable through 1997, but has trended upward significantly since then. Figure 2 reveals that country selection, on balance, offered a wider range of opportunity than sector selection for most of the measurement period, but that the advantage has recently shifted to sectors.
Next, I examine the diversification attributes of countries and sectors. Table 2 compares the average of the correlations between country pairs and between sector pairs, and it reveals a dramatic drop in the diversification advantage of country stratification over sector stratification. This shift in advantage from countries toward sectors occurs not because sector returns have become less correlated, but rather because country correlations have increased substantially in the past few years. Figures 3 and 4 explore this pattern in more detail. Figure 3 compares the correlations of country and sector returns over rolling two-year periods, and Figure 4 plots the differences in these correlations.
These graphs show a convergence of the diversification properties of countries and sectors since the end of 1993, and reveal that as of the past two years sectors have become a stronger diversifier than countries. You might suspect that the relative improvement in sector diversification in recent periods is related to the August 1998 crises when Russia defaulted on its sovereign debt. I re-estimated the correlations excluding the month of August 1998 and found little change in the results. Specifically, for the period January 1997–March 2000, the country correlation dropped from 58.8% to 56.1% while the sector correlation dropped from 63.7% to 59.3%.
Finally, I compute the optimal portfolio given a choice among countries and the optimal portfolio given a choice among sectors. I use the mean returns, standard deviations and correlations estimated from the same four time periods that I used in the correlation experiment. My metric of comparison is the ratio of expected return to risk.
This experiment reveals a pattern similar to the pattern of the previous two experiments. Although there is little difference in the risk-adjusted return opportunities between country and sector optimisation during the period ended in 1991, there is a pronounced shift from a country advantage to a sector advantage from the period ended in 1996 to the most recent period.
The emerging consensus among global investors is that the importance of allocation across countries has diminished in recent periods and that investors should focus more on allocation across global sectors. The evidence seems to support this view:
q The dispersion of returns across countries, on balance, has been greater than the dispersion of returns across sectors, but this difference has shifted noticeably in favour of sectors in recent periods.
q The average correlation among countries has been consistently lower than the average correlation across sectors from 1987 through the mid 1990s, but has declined steadily since then and now favors sectors over countries.
q The risk-adjusted expected return available from optimal allocation across countries versus optimal allocation across sectors has followed a similar pattern. In the most recent period, sectors provided a significant allocation advantage over countries.
These experiments were all conducted from the perspective of a US- based investor. However, I estimated the correlation statistics from the other base currencies as well, and the results tell the same story: the diversification attributes of countries and sectors have converged through time.
I also compared the correlation of countries and sectors during quiet and turbulent regimes, but I did not find any significant distinction as I did between countries and asset classes in earlier research.
Mark Kritzman is managing partner of State Street Associates in Cambridge, Mass. This is a shortened version of an article that appeared in ‘Economics and Portfolio Strategy’, published by Peter L Bernstein Inc in New York