The decrease in equity premium is reasoned by Stulz (1999) and further explained in the paper by Adjaoute et al (2003). They argue that under segmentation, the appropriate measure of risk for a local country portfolio is its standard deviation. When the price per unit of risk is denoted by P, the risk premium on a given security i in the segmented market is therefore , where is the variance of the returns on asset i. In a completely integrated financial market however, investors hold internationally diversified portfolios. The proper measure of risk for the local country portfolio is then no longer its standard deviation, but its beta with the world portfolio. The same asset i in an integrated market will yield a risk premium of where is the beta of asset i. Therefore, if then the risk premium in an integrated market will necessarily be smaller than in a segmented market.
In a broad sense, the EMU represents the highest level of regional economic integration that has ever been reached. The establishment of the European Monetary System (EMS) in 1979 aimed at reducing exchange rate variability, the Single European Act in 1987 which lead to the creation of a free market in the movement of goods, services and capital by 1992, the Treaty of Rome in 1991 opening the way to an economic and monetary union, and finally the foundation of the European Monetary Union in 1999, have established and continuously strengthened real integration among the members of the monetary union.
Moreover, the fact that there is one central bank since 1999 which determines one common monetary policy for all EMU member states, entails one common interest rate and a convergence of inflation rates across EMU member states. The convergence of risk-free real rates, together with the constraints on fiscal policies (member government budget deficits and borrows should not exceed 3% and 60% of national incomes, respectively) and increased cross-border economic activity, are expected to lead to increased synchronization in business cycles across the European economies, resulting in a real convergence of European economies and symmetry of business cycle fluctuations.
There are a number of reasons why this strengthening of monetary and economic integration can be expected to be accompanied by a substantial increase in the financial integration across European markets as well.
First of all, real convergence can have an important effect on financial integration because asset returns reflect to some extent the domestic economic conditions in each country. Having more similar business cycles and being more interdependent through trade could mean a convergence in expected cash flows and volatilities, resulting in a more homogeneous valuation of equities and an increased cross-country correlation in asset prices. Hence, stock markets in EMU member countries should be more integrated as a result of more similar domestic economic conditions across countries after establishment of the EMU.
Secondly, exchange rate fluctuations are mainly driven by national economic policies and also form an important source of risk priced on capital markets. In fact, the existence of exchange rate uncertainty can function as an important device for market segmentation. The more volatile and unpredictable exchange rates are, and the more costly hedging against uncertainty is, the stronger the degree of market segmentation and the lower the degree of correlation across markets. Thus, a more volatile exchange rate of a country raises the national risk premium as investors require a higher return to compensate for the increased uncertainty. Analogously, the reduction or elimination of currency risk, as entailed in EMU and the introduction of the Euro, may raise the degree of financial integration across countries, and simultaneously lead to more homogeneous reward to risk ratios across European stock markets.
Besides these spill over effects from economic and monetary integration, there are also some explicit phenomena that are currently taking place within the EMU that have further intensified and accelerated the path to financial integration within Europe. Recent years have seen positive progress towards financial integration in the EU with the implementation of single market legislation, including the measures of the Financial Services Action Plan (FSAP). This was the result of the consensus agreed by the EU leaders during Lisbon European Council in March 2000. FSAP is an enabling legislation for creating a single financial market, including European prospectus rules and common accounting standards, that would be required to increase cross-border transactions. Furthermore, relaxation of controls on capital movements and foreign exchange transactions, and improvements in computer and communication technology, have lowered the cost of cross-border information flows and financial transactions. Also, more and more companies are now traded simultaneously on the major European exchanges. Moreover, preliminary evidence reports large shifts in portfolio holdings of single member countries away from domestic assets to foreign assets. For example, Adam et al (2002) find overwhelming evidence that the percentage of money invested in funds managed with a Europe-wide investment strategy has made very significant progress since 1999, though this evidence is stronger for money market and bond market funds than for equities. As this phenomenon takes place, risk sharing among member countries should increase and equity markets become more integrated.
To test whether financial markets have truly become more integrated, is to examine whether or not the influence of country-specific risk factors on required stock returns decreases in favour of EU-wide factors. The Capital Asset Pricing Model is used to test for financial integration. In the framework of the CAPM, stock market integration imposes restrictions on the pricing of national assets by ruling out a relationship with purely national risk factors. In completely integrated markets, the only priced risk should be related to systematic risk relative to the EU market. Hence, the expected returns should move in relation to the conditional covariance between the local market and the EU market:
where is the excess return on the local stock market index, is the excess return on the EU stock market index (the universe of investment opportunities is assumed to be the EU market), is the conditional price of the EU market risk, is the conditional covariance operator and is the conditional expectations operator, given information up to time t –1.
In completely segmented markets on the other hand, only domestic risk should affect expected returns. With a single risk factor as a proxy, complete segmentation implies that expected returns are affected solely by the conditional variance of the local market:
where is the conditional local price of risk and is the conditional variance operator given information up to time t-1.
Assuming that markets were partially integrated and the degree of integration varies over time, the conditional mean excess return on the ith stock market index, i=1,…N, can be written as:
where measures the conditional level of integration of market I based on information up to time t –1.
The methodology described above is completely derived from Hardavoulenis et al (1999) but similar models based on the CAPM are used by different authors as well.
Hardouvelis et al (1999) use the above-mentioned model and analyse how a change in the level of integration effects the firm’s equity premium. They model changes in the level of integration by a proxy for the probability of a country joining the single currency. They find that the equity premium in the EU falls gradually as integration increases. The decrease in equity premium is considerable and statistically significant, ranging from an average cumulative effect of 3.6% to 0.5% for sectors and countries alike. Moreover, they find that the equity premium within a sector is indeed converging across EU countries in the sense that its dispersion is decreasing over time. The decrease in dispersion in the same sector across countries is considerable and ranges between 10 and 30% per year. The equity premium is also converging across all sectors in all countries. However, this effect is small (around 4% a year) which reflects the difference in global sector betas. This implies that differences between sectors seem to remain while differences between countries seem to be disappearing.
Other authors have come to the same conclusions as Hardouvelis et al, although using different methodologies. Fratzscher (2001) for example, builds on an uncovered interest parity condition applied to asset prices to define financial market integration, and he employs a trivariate GARCH model for 16 OECD countries to test the implied integration hypothesis empirically. He concludes that European equity markets have become highly integrated with each other since 1996, and second, that the EU market has taken over from the US the role of being the most important market in explaining equity returns for most individual European markets. His paper also compares the relative importance of the three EMU pillars of exchange rate stability, real convergence, and monetary policy convergence in explaining the time variations of equity market integration. He finds that in particular the reduction and elimination of exchange rate volatility and to some extent also monetary policy convergence of interest rates and inflation rates, have been the central driving force behind the financial integration process in Europe.
Yang, Min and Li (2003) investigate the impact of the EMU on the short-run dynamic structure of international market integration, by conducting generalized impulse response analysis and generalized forecast error variance decomposition based on the ECM with imposition of long-run relationships. They conclude that large EMU markets are more integrated with each other after the EMU, while the three smallest EMU markets became more isolated from other EMU markets. Interestingly, they also found that the EMU markets seem to be less integrated with the UK after establishment of the EMU, which provides indirect positive evidence for significant impact of the EMU on European stock market integration. They therefore conclude that the EMU has significantly strengthened stock market integration among its member countries, but lessened linkages with a non-member country (UK) in the same region. This finding is consistent with the argument that macro-economic factors may partly drive stock market movements. However, it is difficult to disentangle the impact of EMU from other channels that might also affect European stock integration. Therefore, the authors themselves put a critical note to this inference, mentioning that the increased integration among European stock markets might also be attributable to faster information transmission and processing due to technological advances, recent consolidation and merger of stock exchanges in Europe etc.
As mentioned before, Yang, Min and Li (2003) also find that the three smallest EMU markets (Austria, Belgium and Ireland) became more isolated from other EMU markets after the EMU launched. Too small a market size and concern of market liquidity is for these markets an obstacle for active participation of international investors, although they have the same macroeconomic linkages with the other EMU countries.
All in all, the studies on financial market integration in Europe seem to indicate that there has been substantial progress in the integration among European stock markets. What is relevant to the research question of this paper however, is the effect of the monetary and financial integration on the stock market correlations among European stock markets. There are some mechanisms through which financial and monetary integration can possibly lead to higher correlations. The coming section will give a conceptual outline on the link between monetary and financial integration and correlations among stock markets.
The effect of monetary and economic integration on correlations is twofold. First of all, it can be expected that due to the convergence of real economies as a result of the increased monetary coordination, the profitabilities of companies across borders will become more interrelated. Another important comovement mechanism is the opening of the market and the increased levels of trade between nations as a result. Consequently, net profits and dividends of European firms, and with that also the stock prices, can be expected to show higher correlation levels across countries.
A second outcome of the economic integration, is the effect of the convergence of monetary policy on the valuation of companies. As inflation rates and interest rates have converged to an European level, dividends and net profitability of companies are discounted at a similar rate, possibly leading to a convergence of stock prices across nations. Furthermore, as exchange rate fluctuations diminish over time, the currency risk factor incorporated in stock prices should be eradicated as well. However, it should be noted that although the adjustment mechanisms in the valuation of stocks may take a long time to take place, the process in itself is in fact a one-time event. Once prices are adjusted to the new macroeconomic variables, there is no reason for them to correlated afterwards in their daily fluctuations. The stock markets would then only comove in a certain direction when a new EU-wide macroeconomic shock occurs (an example of such a shock could be a sudden and sharpe increase of the European interest rate). But in the absence of such a macroeconomic shock, stock markets do not necessarily need to show higher correlations after monetary integration per se.
Similarly, the effects of financial integration on correlations are perhaps less strong than what seems to be an intuitively appealing presumption to make. Due to improvements in computer and communication technology, adjustments in delays in international prices are increasingly shorter and stock markets have become more synchronised. Furthermore, since controls on capital movements and foreign exchange transactions are relaxed, shocks which have an effect on the valuation of many assets worldwide are more easily transferred across countries in an integrated market. For example, after the accounting scandals concerning Ahold, many Dutch institutional investors temporarily lost market confidence and as a reaction, sold a large part of their portfolio, also on a European level. If transaction costs and barriers to trade would have been higher (like before the start of the financial integration process), other financial markets in Europe probably would have been less affected by the market specific shock from the Dutch market.
Thus, as the barriers of the transmission channels in the financial markets are gradually eliminated, one can expect that shocks to common risk factors affect more countries with the same scope, leading to a higher comovement between the markets. But although higher financial integration does provide the means and the channels through which international shocks are transmitted, in itself it does not need to amplify correlations. After all, in a completely integrated market, returns on different assets may be divided into a common component and an idiosyncratic one. The latter however, may be sufficient as to render ex post correlations rather row. Moreover, the common risk component only results in higher correlations when the international risk factor common to all countries is subject to significant fluctuations and shocks. As a conclusion, it can be said that higher financial integration in itself does not necessarily result in higher correlations. In fact, higher correlations are neither a necessary nor sufficient condition for higher market integration (Adler and Dumas, 1983).
4.3. Country versus industry effects:
The theoretical reasoning and empirical studies mentioned above seem to indicate that financial market integration in Europe is not only quite probable but is in fact already taking place, although it remains difficult to identify the exact character of the driving forces behind it. Most probably however, the convergence of a common monetary and economic policy in the EU has reduced the impact of country specific shocks and has puts a greater importance on EU wide shocks that affect all stock indexes across the mainland. This, together with the diminishing of the home bias effect and the reduction of segmented market as a consequence of financial integration, does not give much hope on finding low correlations among EMU country returns, and weakens the case of international diversification among EMU countries.
So far, we have examined cross-country correlation levels to determine whether diversification among EU national indices could be profitable. However, analysing the evolution of the average correlations across countries and industries index returns may be a rather naïve analysis.
As is argued by Beckers at al (1992), low correlations between different markets may be perfectly consistent with complete integration. Different industries may play a crucial role in each market, for example pharmaceuticals in Switzerland, oil in the UK, forestry and paper on Finland etc. Stocks within a given industry may be similarly impacted by global events: oil companies around the world in general should react in a similar way to changes in oil prices or the invasion in Iraq, financial institutions may broadly be affected in the same way by the overall level of interest rates regardless of the EMU country they are situated in, textile companies may react similarly to GATT decisions on production and import quota etc. Random shocks which selectively affect different industries would therefore naturally lead to low correlations among nations when particular industries are more present in one country than another.
Also, looking at average correlations only, ignores the differences between indices composition across markets. For instance, Ferreira (2003) points out that Finland and Belgium index returns have a correlation of 0.24 in the 1999-2001 period. However, this figure does not tell us whether this relatively low correlation level is due to the ‘national’ differences between the two countries, or whether it is due to the fact that the Finland index return has a Information Technology industry weight of 44% while Belgium’s has an inexpressive 0.1%. Thus, the evolution of the correlation between two indices could be caused either by the industry or the country factors of each index return.
In order to properly measure the relative importance of country and industry factors in explaining return correlations it is necessary to decompose the return on a give stock or industry index into a common, country and an industry effect. Therefore, a simple factor model proposed by Heston and Rouwenhorst (1994) is often used to measure the relative importance of country and industry factors in explaining return correlations. The return on a given stock or industry index is then decomposed into a common factor, a country effect, an industry effect and a firm-specific disturbance.
Rit= αt + βjt + γkt + εit
where Rit is the return on the stock i at time t, αt is the common factor, βjt is industry j effect, γkt is country k effect and εit is the error term.
If it turns out that country factors dominate industry factors in explaining the low correlations, then that would imply that for example, the occurrence of shocks that affect banks in Switzerland differently from banks in Sweden is more important for explaining the low correlation between their country returns than the fact that Sweden has fewer banks. Or alternatively, cross-country variation in investor sentiment drives a wedge between the return of firms that are in the same sector but located in different countries. This indicates that differences in institutions across countries affect the transmission of global shocks to asset values. Differences in interest sensitivity of aggregate demand (due to for example heterogeneous industrial structures), the importance of the ‘credit channel’ (differences in the structure of the banking system, in business firms’ reliance on bank loans, and in financial stability) are all potential determinants of the country-specific economic effect of global shocks (Mihov (2001)). As such, even if a common monetary policy is carried out, as long the transmission of monetary policy to economic activity remains different among EMU countries, business cycles may not synchronise and correlations among EMU country indices may remain low.
Until recently, most of the literature found an overall dominance of country effects over industry and common market effects. For example, Beckers et al (1992, 1998) have shown that differences in the industrial makeup of countries only plays a minor role in explaining country correlations. Instead the low correlations are primarily due to large country-specific sources of return variation. Also, Heston and Rouwenhorst (1994) reconfirmed the conclusions of King (1966) and Lessard (1974), stating that little of the variation in country index returns has been explained by their industrial composition. They find that the ratio of the cross-country average variance of the pure country effect (24.32%) to the cross-industry average variance of the pure industry effect (6.46%) was 4.5. This predominance of country versus industry effects is also found in subsequent empirical work of Griffin and Karaloyi (1998), Heckman, Naryanan and Patel (1998).
The trend, however, appears to have reversed lately, as papers using very recent datasets have detected an increase in the global industry effects. For example, Cavaglia, Brightman and Aked (2000) working with a sample period over 1986 to 2000, showed the increasing importance of industry effects relative to country effects.
But although most authors agree that the role of industry factors has been increasing in the last decade, there is still not a consensus on the dominance of either the industry of country factor after the introduction of the euro. Some recent research namely found that industry effects are even higher than country effects during the most recent sample periods, after 1995. For example, Isakov and Sonney (2002) obtained a country/industry effects ratio of 1.6 using a sample of 20 developed countries over the 1997-2000 period. In addition, they found a predominance of the industry factors using a sub-sample of 9 European countries. Also, Rouwenhorst, after extension of his sample size to July 2000, recently finds that the importance of industry effects has risen sharply since 1998 and that industry effects are nowadays at least as importance as country effects.
On the other hand, Carrieri, Erruza & Sarkissian (2000), Gerard, Hillion & de Roon (2002), and Adjaoute & Danthine (2001a, 2001b) also find the domination of country effect has diminished, but they conclude that industry factors are still less important than country factors. Ferreira (2003) for example finds a ratio of 1.130 which indicates that country factors still rule over industry and common effects.
Brooks & Del Negro (2002a) conclude that the rise of the industry effects is less pronounced when the sectors of the IT bubble (Technology, Media and Telecommunication, TMT afterwards) are neglected. The absence of evidence for industry factor dominance beyond TMBT sectors and the US is interpreted by the authors as an indication that the recent dominance of industry effects over country effects is a temporary phenomenon associated with the stock market bubble.
In Brooks & Del Negro (2002c), they criticise the methodology of Heston & Rouwenhorst (1994) which is described above and followed by most papers. The dummy approach followed by the HR methodology assumes that all companies are a member of exactly one country and one industry. Using an unrestricted model without this strong assumption, they find that country factors are more important in explaining stock return. However, they (Brooks & Del Negro (2002b)) report that Europe (unlike other regions in the world) is the only region where the relative and absolute importance of the industry effect has increased, even when using the adjusted Heston & Rouwenhorst methodology, and even when the TMT sectors are excluded.
4.4. Analysis of mean-variance frontiers
The results found in the literature on the importance of industry effects versus country effects in explaining stock returns are thus mixed and ambiguous. And although industry and country effects are important factors shaping the correlation structure among European stock markets, the theoretical discussion described above is not necessarily relevant to European investors in practice. In the end, assuming that investors invest in the general market index of a country, then diversification benefits can still be exploited when one spreads his portfolio over two countries which are weakly correlated, regardless of the country/industry factors behind the correlating structure. After all, when a particular sector would have dominated both markets (so industry factors were indeed more important), then correlation would have been high and there would have been no incentive to diversify over both markets in the first place. Thus, regardless of the underlying factors causing correlations among country indices, the implications for an investor in terms of diversification opportunities are still the same. What the theory of industry/country effects does tell us however, is that caution must be made not to invest in the same sector in different countries when a geographical diversification strategy is chosen. Obviously, such a policy would limit the advantages of the diversification strategy as the same sector can be hit by a global shock that has similar effects in both countries. But in other aspects, the importance of industry versus country effects to investors is rather limited, since regardless of the underlying causes, the implications for a mean-variance optimising investor are the same
Although the time-varying correlation among country and industry indices, and the relative importance of industry effects to country effects on stock returns has been extensively investigated, none of these studies really touches on the heart of the matter. The primary goal of all this research would be to find the optimal asset allocation strategies for a portfolio manager in Europe, but only very few papers directly address this essential question. In fact, I have only found two studies that have directly documented the effects of integrating Europe on investment strategies; the study by Ehling and Ramos (2002) and Moerman (2003). In order to answer the central question, both of these studies advocate an alternative, more direct research design from the ones employed in the past, namely looking at portfolios instead of at country or industry effects.
For this purpose, the portfolio optimisation problem from the point of view of an investor with a mean-variance perspective is solved (based on the theory of Markowitz (1952) discussed above). Subsequently, both papers use a spanning test to statistically test their results, although Moerman. The principle of the mean-variance spanning test was however first initiated by Hubermann and Kandel (1987). The intuition behind the spanning test model is perhaps best explained by Moerman (2003). He argues that since from a mean-variance point of view, adding assets to the current portfolio will lead per definition to a shift of the efficient frontier, a spanning test needs to be performed in order to see whether this shift is statistically significant. The spanning test compares the whole set of efficient portfolios and tests whether the addition of the other set of indices (eg adding industries to a current portfolio of country-indices) gives significantly better portfolios. A rejection of the spanning test means that the country indices do not span the universe of both types of indices; the shift of the efficient frontier is statistically significant and so every investor is better off considering both investment categories.
I will now continue to compare the findings of both studies.
Ehling and Ramos (2002) specifically test whether the advent of the EMU really has an impact on portfolio strategies. In order to do so, they try to isolate the EMU effect by considering four different samples: 11 EMU countries, 3 European Community (EC) but non EMU countries, 3 candidate countries who are in the process of future entrance to the EC, and 2 other European countries. For all samples, they use weekly, euro denominated DataStream data.
They find that in the pre-convergence (1988-1994) and convergence period (1995-1998), industries underperformed countries, i.e. overall industry portfolios tend to be located below the security market line constructed out of country indices. In the euro period however, the traditional advantage of country portfolios over industry portfolios seems to disappear, since they both have the same mean (the slope of the tangency portfolios starting from the origin to the minimum-variance frontier is identical for both strategies; please refer to section 3 for explanation of the mean-variance terminology ).
However, the above mentioned findings depend heavily on the assumption that investors hold a portfolio close to the tangency portfolio. If investors would be interested in holding the minimum-variance portfolio, they find that investing in both country indices as well as in industry indices, would be the optimal diversification strategy. This conclusion is based on their spanning tests which show that neither country portfolios span industry portfolios nor industry portfolios span country portfolios. This implies that in the past, country motivated portfolios provided diversification benefits over industry portfolios, if any, only in the case investment took place close to the tangency portfolios starting from the origin (Ehling and Ramos (2002)). Furthermore, they also find that with short sale restrictions, the outperformance of countries over industries is reduced.
Surprisingly, they cannot find any evidence of the advent of the EMU having a significant impact on their results, since all results are identical across the groups of countries (EMU, EC and candidate EC countries). Their indefinite conclusion is that either the EMU is not directly responsible for the apparent shift in the mean returns, or it has affected all the countries in Europe regardless of whether they had already joined the EMU or not.
The study by Moerman (2003) uses approximately the same methodology as Ehling & Ramos (2002), although the scope of his research in terms of sample size (both number of countries and number of observations) is less comprehensive. Moerman addresses the issue of diversification within European capital markets for both short term and long-term oriented investors by studying both unconditional and conditional mean-variance frontiers. He uses monthly returns as derived from MSCI indices for 10 EMU-participating countries and 10 EMU industry indices from January 1995 – October 2002. His dataset is split into two subsamples; January 1995 until December 1998, and January 1999 until October 2002.
Both the long-term and short-term analysis support his initial conjecture that the performance of a pure country investment strategy is decreasing. The unconditional analysis (for the long term investor) until January 1999 shows that a pure country-allocation strategy resulted in a similar performance as a pure-industry allocation strategy scheme. Since he finds that neither the country indices nor the industry indices span the mean-variance frontier for both types of investments, he concludes that during the first subsample period, the best portfolio can be constructed when the investor considers both categories simultaneously.
However, he finds that in the period January 1999 - October 2002, a more efficient portfolio can be constructed using industry indices compared to using country indices only. Also, he cannot find support in favour of adding country indices given a mean-variance efficient industry index allocated portfolio. Also for the conditional model (short term investor) in the second subsample, he finds that diversifying over industries is a much better strategy than diversifying over countries. He argues that shift in optimal asset allocation strategy is a consequence of the harmonisation in the monetary and policy rules in the European Monetary Union.
Although both studies report an increasing gain of importance for industry asset allocation strategies, the contrast in the results found by both studies are remarkable and noteworthy. While Ehling and Ramos find that in the pre-euro periods, pure country-allocated portfolios performed better than pure industry-allocated portfolios, Moerman finds that the performance of industry portfolios was equal to country portfolios. And where Ehling and Ramos find that from 1999 onwards, countries and industries have similar mean-variance optimising benefits, Moerman finds that industry portfolios outperform country portfolios. Furthermore, while Ehling and Ramos find that mixing both strategies offers diversification gains during all sample periods, the paper of Moerman rejects this finding after the introduction of the euro in favour of a pure industry-allocated portfolio. This last finding in particular is striking, since it would be a clear indication that industry indices are more important than investing in country indices nowadays.
Since both studies use the same methodology, the divergence of the conclusions reached are most probably the result of the different data sources employed. Where Ehling and Ramos use DataStream indices (starting from January 1988 till December 2001), Moerman makes use of MSCI indices (starting from January 1995 till October 2002). Data from the Morgan Stanley Capital International (MSCI) indices are based on high-capitalisation companies of the markets, while DataStream indices are defined as value-weighted broad indices of national stock markets, covering also medium- and small-capitalisation companies.
The drawback of using the MSCI datasource for measuring stock market integration and segmentation is that large firms are likely to be more effected by EU factors given that they are more likely to be engaged in international operations than small firms (Hardouvelis et all, 1999). Consequently, their stock prices may be more sensitive to EU factors than small firms’ stock prices are. As a result, using indices that omit small firms (like the MSCI indices employed by Moerman) may bias the results towards finding integrated markets and higher correlations among country indices. This might be a possible explanation for the relative importance of pure industry-allocated portfolios found in the study by Moerman.
One last striking difference between the studies is the fact that Moerman (2003) states that the gain in the importance of industry asset allocation strategies are due to the integration process within the European Monetary Union, without giving further evidence for this claim. Ehling and Ramos (2003) however, explicitly investigate this issue, and are unable to conclude that the EMU plays any causal role for diversification strategies. Since the assumption made by Moerman is not supported by further empirical evidence, his conclusion on this particular issue seems somewhat preconceived.
5. ADDITION TO CURRENT BODY OF KNOWLEDGE:
As can be noted from the literature review above, the extent to which financial markets have become more integrated has been the topic of extensive debate. Most papers however, focus on the importance of country versus industry effects in explaining stock returns. Only two studies so far have been published which investigate the direct implications for the mean-variance optimising investor, both with somewhat diverse results, leaving the reader still inconclusive.
As indicated above, the main problem of the study by Moerman (2003) may be that his conclusions may have been biased a priori towards sector allocation strategies, because the data source he employed (MSCI indices) is based on high-capitalisation companies which are more likely to be effected by EU factors. And although Ehling and Ramos (2002) use a more appropriate datasource, their sample period for the euro-period is rather short; only two years.
Currently, we have more than five years of data available for the post-euro period. It would therefore be interesting to contribute to the discussion by extending the scope of the methodology of Ehling and Ramos (2002) and Moerman (2003) by conducting the research over a longer sample period. Furthermore, Datastream indices will be used to avoid the problem of high-capitalisation bias.
Moreover, in order to capture the process of integration in the European stock markets, I perform the analysis over four subperiods, each indicating a different phase of economic integration in the European Union: the pre-convergence, convergence, pre-euro and euro era. An assessment of the optimal diversification strategy for each subperiod will be made to evaluate the consequences of the European integration process for the mean-variance optimising investor.
By extending the sample period using more recent observations, and by overcoming the problem of high-capitalisation bias, I hope to be able to reconcile the findings of both studies and to reach a more unambiguous conclusion on the question of the optimal diversification strategies within the EMU.
6. METHODOLOGY AND DATA:
6.1 Methodology
From the general research questions composed at the beginning of this research, and on the basis of the above mentioned literature, I have drawn a number of hypothesis which I would like to investigate further.
Hypothesis 1: Correlations between EMU country indices are constant across the different subperiods.
An increase in the correlation coefficients could indicate increased market integration over time, and a possible reduction of diversification benefits for a mean-variance optimising investor. Therefore, a comparison is made of the correlation coefficients among the country and industry indices over time to see whether there are significant changes in the four subperiods specified.
To illustrate that the increase in average correlation trends is not heavily influenced by the figures of a few countries only, the following graphical analysis is conducted following Adjaoute and Danthine (2001). First, the country or sector index returns are used to compute the unconditional correlation matrices for the different periods. Secondly, the correlation pairs are sorted in ascending order and plotted against the country pairs for each subperiod (eg pre-convergence period). Then the correlation pairs for the subsequent period (eg convergence period) are recomputed and plotted along the correlation pairs of the previous period (pre-convergence) correlation pairs.
Furthermore, the change in the correlation over time will statistical be tested statistically, by comparing the outcome of to the critical value of 1.96 (5% significance level). In this test, and stand for the average correlation in two sequential subperiods, and T is the number of observations within each subperiod.
Hypothesis 2: Pure-country diversification in the EMU market leads to significantly better risk-return tradeoffs than a pure-industry diversification strategy.
For this purpose, the portfolio optimisation problem will be solved following Markowitz (1952) who solves the problem from the point of view of an investor with a mean-variance perspective. Given the means – which are estimated by the historical averages – and the covariance matrix of the country and industry indices, the efficient mean-variance frontiers of each will be plotted.
In the standard Markowitz (1952) the investor minimizes the amount of risk of his portfolio as measured by the variance given a certain demanded return and subject to the budget restriction that all weights should add up to one. The unconditional portfolio optimisation problem that needs to be solved is therefore the following:
The optimisation problem thus requires finding the portfolio weights w1, w2…wn that minimise the portfolio’s risk (as measured by the portfolio variance), given an expected return and a budget restriction of a fully invested portfolio in which all weights should sum up to one.
In contrast to Ehling and Ramos (2002), who use a partial spanning test to determine which asset allocation strategy is optimal in the tangency portfolio point, this paper has deliberately chosen to compare Sharpe ratios. Although the spanning test tests whether the mean returns of the alternative investment strategies are equal or not, it does not test the equality of the standardised returns in the tangency portfolios. Since it is well known that investors make a trade-off between both return and risk, this paper has chosen to test the equality of the reward-to-variability ration of the tangency portfolios. I will assess the risk-return trade-off by using the reward-to-variability ration, or more popularly referred to as the Sharpe ratio. It is defined by
where E (RP) denotes the expected return of the portfolio; RF denotes the return on the risk-free asset; and σ (RP) denotes the standard deviation of the portfolio returns. This ratio measures the excess return, or risk premium, of a portfolio compared with the risk free rate, compared with the total risk of the portfolio, measured by its standard deviation. The average risk free rate per subperiod is calculated using the Datastream weekly returns of the Dutch government bond index.
The Sharpe ratios of the tangency portfolio’s of a pure-country diversified strategy, a pure-industry diversified strategy, and a strategy which allows for diversification over both asset classes will be compared. A statistical test (2*sqrt(2/T)) will be used to test whether a certain Sharpe ratio is significantly larger than the Sharpe ratio of another asset class.
Hypothesis 3: An investor can significantly improve his portfolio by investing in both industry and country indices.
For this purpose, the alternative set of indices is introduced to the optimal diversification portfolio (either pure-country or pure-industry diversified) to see whether the investor can significantly improve his portfolio by investing simultaneously in both indices. Since from a mean-variance point of view, adding assets to the current portfolio will lead per definition to a shift of the efficient frontier, a spanning test and an intersection test needs to be performed in order to see whether this shift is statistically significant.
The principle of mean-variance spanning is based on a set of K benchmark assets and a set of N test assets. The K assets span a larger set of N+K assets if the minimum-variance frontier of
the K assets is identical to the minimum variance frontier of the N+K assets. The spanning test which is employed here is based on the recognition that investors can construct their portfolio either entirely out of country indices or industry indices. We therefore have two sets of benchmark assets and two sets of test assets. Two types of regressions are then carried out:
Rcoun,t+1 = αcoun + βcoun R ind,t+1+εcoun,t+1
Rind,t+1 = αind + βind R coun,t+1+εind,t+1
Where Rind,t+1 is a K·1 vector of industry index returns for time t, and Rcoun,t+1 is the L·1 vector of country index returns for time t. Furthermore, αind (a K·1 vector of constants) βind (a K·L vector of slope coefficients), αcoun and βcoun are parameters to be estimated.
An intersection test is performed to test specifically the case of someone holding the tangency portfolio of one investment set. Given a specific interest rate (η) and a specific point on the efficient frontier (the tangency portfolio), the intersection test tests whether mean-variance investors can improve their mean-variance efficient set by including the other set of indices.
The hypothesis of the intersection test can be stated by the following restrictions:
αcoun - η· (ιind - β · ιcoun) = 0
were ιcoun and ιind are K·1 and L·1 unit vectors respectively with all elements equal to one. The intersection is thus a Wald-test of K restrictions at the same time, where K is the number of new investment opportunities introduced. η can be seen as the interest rate, but since excess
returns will be used for the intersection test, η will be set to 0 in this paper. The null hypothesis is that the mean-variance frontier generated by the countries alone intersects the frontier generated by the larger set of assets, which include the test asset (so countries and assets), at a tangency point corresponding to the mean of the Dutch risk free interest rate.
The more intuitive methodology behind the intersection test is the following. First the optimal tangency portfolio is calculated for two cases: one for the benchmark asset class (either industries or countries), and one for the benchmark plus the test asset (so both industries and countries). The test then compares the tangency portfolios and test whether they are “close” to each other in mean-variance space or not. If it is found that the two tangency portfolios are close to each other (so the lines going through the interest free rate to the tangency portfolios of both cases intersect), then one could conclude that adding the test asset to the universe does not significantly shift the mean-variance frontier outwards. In that case, it would be sufficient to invest in countries only. This case is illustrated in Figure 2A, where the solid curve is the mean-variance frontier generated by a country portfolio, the dashed line is the frontier generated by expanded set of assets in both countries and sectors and the intercept of the tangency line is equal to Rf. Figure 2B on the other hand illustrates a situation when the two tangency lines do not intersect. If the intersection test is then indeed rejected, it is proven statistically that the slope of the two lines is significantly different, and so investors holding the tangency portfolio are better off investing in both industries and countries.
Figure 2A: The Case of Intersection
Figure 2B: The case of no intersection:
Spanning refers to the situation where there is intersection for all risk aversions (all possible interest rates). The restrictions of the null hypothesis of the spanning test (as conducted by Moerman(2003), based on the test of DeRoon and Nijman (2001)) can be stated as follows
αcoun =0 and β · ιcoun - ιind =0
were ιcoun and ιind are K·1 and L·1 unit vectors respectively with all elements equal to one. The test consists of 2*K restrictions and the Wald-statistic is distributed with 2*K degrees of freedom. From this we can also see that the spanning test is more restrictive than the intersection test: if the restrictions of the spanning test hold, by definition the restrictions of the intersection test hold as well.
In fact, if the restrictions of the intersection hold for two arbitrary chosen interest rates, it is said that the frontier generated by country assets spans the frontier generated by the larger asset set (i.e. the two frontiers coincide at every point). The fact that we only have to test for two different values of can be explained by well-known results from portfolio theory. From mean-variance mathematics we know that every portfolio on the frontier can be obtained as a linear combination of two distinct portfolios on the frontier. Hence intersection at two points on the frontier is equivalent to intersection at all points on the frontier, which is the definition of spanning (Torbjörn Sällström, 1999).
The case of spanning is illustrated in Figure 3. As explained above, panning refers to the case where the whole mean-variance frontier generated by an expanded investment opportunity set of benchmark assets and some test assets coincides with the frontier generated by the subset of benchmark assets only. A spanning test thus compares the whole set of efficient portfolios to see whether one can be replicated by the other; it is by definition the situation that intersection holds for all interest rates. Like with the intersection test, if the entire null hypothesis of the spanning test is rejected, we can conclude that the country indices do not span the universe of both types of indices, and so investors should form their portfolios out of both country and industry indices.
Figure 3: The case of spanning:
Following Kan and Zhou (2001), Ehling and Ramos (2003) split the null hypothesis into two parts, conducing two separate tests, employing a socalled step-down approach. If αcoun =0 and β · ιcoun - ιind =0, we can conclude that investors should base their strategies on industry portfolios only, since industries span countries. On the other hand, αind = 0 and β · ιind - ιcoun =0 implies the opposite, that is, mean variance spanning is not rejected and thus country allocation is the superior portfolio diversification strategy. If both tests are rejected, then spanning does not occur, and then an investor is better off investing in both indices. Ehling and Ramos argue that the step-down approach has the advantage that in case spanning is rejected, one knows exactly if the rejection originates from a difference in the means between the test asset and the benchmark assets, that is, the slope of the tangency portfolios starting from the origin is incongruent, or if the standard deviation of the minimum variance portfolios are not the same. However, in their paper, it is apparent that although the first part of the test is sometimes accepted, the second part of the test (β · ιcoun - ιind =0) is always rejected when spanning does not occur. Furthermore, since equality of mean returns in itself is not so important for investors, this paper has prefers to compare standardised mean returns (Sharpe ratios instead of splitting the spanning test (please refer to see hypothesis 2). The further conclusions from both variations of the spanning test should be the same however, and therefore, this paper has chosen to conduct the simultaneous test as proposed by DeRoon and Nijman (2001) and followed by Moerman (2003).
6.2 Data:
This study employs total market indices composed by DataStream International. There are three specific reasons why this study has chosen to employ DataStream as a source of data.
First of all, Datastream indices are expected to be more homogeneously composed across markets than the national market indices, hence making empirical results more comparable over different countries.
Secondly, Datasteam indices are preferred to the widely-used Morgan Stanley Capital International (MSCI) indices since they are defined as value-weighted broad indices of national stock markets, covering also medium- and small-capitalisation companies. As such, they are more likely to proxy the whole equity market as opposed to indices based on high-capitalisation companies such as the MSCI index. The difference is likely to be important when addressing issues such as stock market integration and segmentation, since small-capitalisation stocks may behave differently than large capitalisation stocks.
Thirdly, the definition of industries is consistent for all countries of the Datastream database, which eliminates the potential bias induced by misclassification of firms (Ehling and Ramos (2002)).
Weekly, dividend adjusted continuously compounded stock returns will be used for 12 countries that have signed up for EMU: Austria, Belgium, Finland, France, Germany, Ireland, Italy, Netherlands, Portugal, Greece, Luxembourg and Spain. Industry analysis is based on Datastream sector classifications at level 3 consisting of 10 sectors: Basic Industries, General Industries, Cyclical Consumer Goods (CCG), Non-Cyclical Consumer Goods (NCCG), Cyclical Services, Non-Cyclical Services, Utilities, Information Technology (IT) and Financials.
The sample period is from 1979:3 to 2004:3. In order to capture the process of integration in the European stock markets, I wish to perform the analysis over four subperiods, each indicating a different phase of economic integration in the European Union:
Period 1 Jan 1979 - Dec 1985: The pre-convergence period.
Developments following the establishment of the European Monetary System in 1979 aimed at reducing the fluctuation of intra-European exchange rates and at safeguarding price stability
Period 2 1 Jan 1986 – Dec 1993: The convergence period.
In February 1986 , the Single European Act imposed a deadline, the end of 1992, for the complete abolition of all obstacles to free cross-border movement of labour, capital, goods and services. At the same time, the regulatory structure of the financial markets was harmonised with the First and Second Banking Directive and a set of other legislative and institutional measures.
Period 3 Jan 1994 - Dec 1998: The pre-euro period.
In February 1992, the Maastricht Treaty was signed, which became effective in November 1993, and opened the possibility for a monetary union among those EU countries that would satisfy the convergence criteria on inflation, interest rates, the public finances and exchange rates.
Period 4 Jan 1999 – April 2004: The euro period.
In 1999, a common monetary policy conducted by the European Central Bank became effective across EMU and the common currency, the euro was introduced in giral form.
I believe that these subperiods are representative for a particular era in the process of the integration of the European Monetary Union. Consequently, the statistics on the returns, variances and covariances of the local country indices during these periods should reveal important changes over time.
Unfortunately, data is not available for Finland, Portugal, Spain, Greece and Luxembourg for the pre-convergence and convergence period. Therefore, data from these countries will only be taken into consideration for the pre-euro and euro era.
The price series used in the study will be transformed by taking natural logarithms of the raw data. For further ease of comparison, all returns will be expressed in the Dutch guilder until 1st January 1999, and afterwards in the common euro currency. This method has the preference above the expression in local returns, since in practice most investors do not hedge their returns and are subject to exchange risk fluctuations. However, expressing returns in a common currency may introduce a bias towards higher correlations of returns over the years when the EMU exchange rates have become more stabilised and correlated. This makes it more difficult to draw conclusions from the evolution of the correlation matrices, since higher observed correlations can also be caused due to financial integration as well as a convergence of exchange rate fluctuations. The use of local currency returns are justified however, since irrespective of the cause of the observed development of correlations over time, the practical implications to local investors are the same. And since this study has deliberately chosen to emphasise and utilise a practical approach, the expression of returns in the Dutch guilder is justified.
- MAIN EMPIRICAL RESULTS:
7.1 Descriptive Statistics
Table 1 shows summary statistics (average annual returns, annual standard deviations and corresponding Sharpe ratios) over the different sub-periods, for both countries and sectors. It should be noted that the reward-to-risk theory (from now on referred to as the Sharpe ratio) is based on expected returns. In subsequent chapters, the historical return over the whole 35 year-old sample period will be used as a proxy for the ex ante expected return. However, for the descriptive statistics of this chapter, the Sharpe ratio’s are calculated using the average returns per subperiod. Due to a lack of data, no information is provided for Spain, Greece, Finland, Luxembourg and Portugal during the first two subperiods.
The table illustrates a number of things. First of all, the table clearly shows that the mean returns in particular went through different phases during the different subperiods. For example, the average mean return of countries for the pre-convergence period was 18.3%, for the convergence period 10.4%, for the pre-euro period it was 19.1% and during the euro-period it was only 0.6%.
Although standard deviations varied as well over time, the divergence is much smaller. As a result we can see that the standardised returns (Sharpe ratios) vary considerably through time as well. Correspondingly, during the pre-euro period, almost all stock markets in the world experienced a dramatic increase in value, only to deflate just as sharply a few years later during the euro era. Dividing the sample into four subperiods is therefore not only a good strategy because certain periods are associated with important institutional and structural changes occurred in Europe, but also because the mean return has varied considerably from one period to another. This observation also makes us realise that looking at the relative position of the mean-variance frontiers for countries in different subperiods is pointless. Comparing the diversification benefits of one asset allocation strategy (countries or sectors) with itself through time is incorrect, since the position of the mean-variance frontier is very much influenced by the investment climate in the different time periods.
The second thing which becomes visible from Table 1, is the fact the average returns, standard deviations and Sharpe ratios over the sample periods are very similar for countries and sectors alike. That is, both asset classes are heavily influenced by the investment climate, and shift jointly during the different economic cycles. Looking at the whole sample period though, we see that the industry indices outperform country indices in terms of standardised return (0.74 versus 0.64). The difference in Sharpe ratios is significant at a 5% level, and this difference can be completely attributed to the lower standard deviation of industry indices.
When we look within the asset classes in more detail, we see large differences between the individual assets. Ireland has the best absolute performance for countries (average annual return of 13.45), but when corrected for its high average standard deviation, it does not have the best standardised return. The best standard return for countries is achieved by the Dutch index with a Sharpe ratio of 0.88, and by the Non-Cyclical Consumer Goods among the industry sectors (Sharpe ratio of 1.00). The best absolute performance in industries is gained by the Information Technology sector, with a average return of 15.7% for the whole sample period. However, since the IT sector also has the highest standard deviation of all assets considered, it ends up with the lowest Sharpe Ratio of the entire industry sample.
We can also see from the table that the relative performance of the different assets over the sample period is very volatile. To illustrate, Table2 contains the best and worst performers of country and industry indices over all the subsample periods, and is directly derived from Table1. There is a large difference between the standardised returns of the best performing and worst performing assets. Furthermore, the countries and sectors that show extreme results differ in each subperiod. We see for example that Austria, who was the worst performer in the pre-euro period, outperformed all other indices in the consequent euro-period. One thing which is striking though, is that the Netherlands has the best performing stock market in all the subperiods except for the euro period. Likewise, Resources outperforms all other industry indices in two of the four subperiods.
7.2 Correlations:
This chapter will first give a description of the correlation structure of the assets in the period January 1979 until April 2004. In the second section of this chapter, these empirical observations are discussed and put into perspective by making a comparison to the findings of other studies, and by relating the results of this study to the theory of financial and monetary integration.
7.2.1. Description of results:
Table 3 shows the average correlations across countries and industries index returns. Columns A of each period consists of the average cross-correlation of an asset with the other indices from the same asset class (eg the average correlation of the Dutch stock index with the other 11 EMU countries). Columns B consistent of the average cross-correlation of an asset with the indices from the other asset class (eg the average correlation of the Dutch stock index with the industry sectors). Unfortunately, again no analysis can be made for the indices from Spain, Greece, Finland, Luxembourg and Portugal in the pre-convergence and convergence subsample.
We will first concentrate on the columns A of each sample subperiod. The first thing that is notable is the fact that average correlations among sectors (0.560) are substantially higher than country correlations (0.396). This is a bad sign for diversification strategies among industry sectors. However, as is indicated by Ferreira (2003), such results should be carefully interpreted as industry indices are biased by the weight of some countries in their composition.
The results for the correlation coefficients depicted here reconfirm the findings of Yan, Min and Li (2003) who conclude that smaller European markets are less integrated with the rest of the European capital markets, since their small market and concerns of market illiquidity most probably withhold investors from investing in those markets. Indeed, it can be derived from Table 1 that the average correlation coefficients for countries with a smaller capital market (eg Austria, Greece, Ireland and Luxembourg) with the rest of the EMU stock markets (0.32) are much lower than average correlations of the larger markets (0.43). Italy is the exception in this trend since that is the only country with a large market capitalisation while at the same time experiencing relatively low correlations with the rest of the EMU countries until the last subsample period.
Table 3 also clearly depicts that correlations have increased considerably through time in order to reach a maximum in the 1995-1998 period, and then decrease slightly in the last sample period. Table 4 shows the average increase of the asset class correlations over different sample periods. Please note that the average increases cannot be derived from the correlations means reported in Table 3, but are directly calculated from the original values of the correlation matrices to produce refined results. Table 4 outlines a drastic increase of correlations in the convergence period; the mean difference between correlations coefficients in the pre-convergence and convergence periods was 0.306 points on average. The drop in the Euro period (-0.081 points) is relatively small, and less uniform.
Although both countries and industries experienced an upsurge in correlations, Table 4 shows that the increase over all subsample periods has been the highest for countries (total increase of 0,469 points for the whole sample period). The difference of total increase with for example the total correlation increase of sectors, mostly comes from the divergence in the pre-euro period, where correlations among countries increased much more than among sectors or countries§ors (0.173 versus 0.036 and 0.036), and from the euro period, where the average decrease was less explicit than with the other classes. (-0.027 versus –0.151 and –0.067 respectively).
The finding that the average increase of correlations is highest within country indices, is also found by Rouwenhorst (1999), and is consistent with the assumption that integration within Europe has the largest impact on country indices. This might be a sign of decreasing diversification benefits within country indices relative to sector indices.
To illustrate that the change in average correlation trends as described above is not heavily influenced by the figures of a few countries only, a graphical analysis is conducted following Adjaoute and Danthine (2001) where the correlation pairs are sorted in ascending order and plotted against the country pairs for each subperiod (eg pre-convergence period). Then the correlation pairs for the subsequent period (eg convergence period) are recomputed and plotted along the correlation pairs of the previous period (pre-convergence) correlation pairs.
The results are depicted in figures 4A to F. Figure 4A shows the evolution of country pair correlations by comparing the pre-convergence and convergence period. It is obvious that correlations between all country index pairs were uniformly larger in the second period than in the first. The average increase of the correlation coefficient was remarkable: 0.323 points (please refer to Table 4). Figure 4B shows a similar pattern for the convergence and pre-euro period. Again, coefficients for all the correlation pairs are larger in the second period than in the first, although the average increase is less impressive: 0.173 points. Figure 4C however, shows a radically different picture. In the euro period, it is no longer the case that all correlations are uniformly higher than in the immediately preceding period. In fact, 37 out of 65 correlation coefficients are lower in the euro sample than the pre-euro sample. The average decrease in correlations was a small 0.03 points, though still significant at the 5% level.
Figures 1D to 1F depict the same methodology applied for the industry sectors. We see that also for sectors, correlations in the convergence period were a lot higher than during the pre-convergence period. But although substantial, the 0.286 points of average increase in correlations is lower than the country correlation increase. Two other things are notable from Figure D. First of all, comparing Figure D with Figure A, it becomes apparent that indeed correlations among the industry indices are generally much higher than among country indices. The second that is striking about Figure D, is the convergence of the two lines plotted. It is striking to see that those sector pairs which had the lowest correlations in the pre-convergence period show the largest increase of correlations in the convergence period. On the other hand, sector pairs which were already highly correlated with each other in the pre-convergence period experienced only a relatively small increase of correlations in the subsequent period. A similar pattern can also be detected for the country pair correlations, although the convergence effect for countries is generally smaller. However, where for sectors, the range in correlation levels (the difference between highest and lowest correlation per period) is seen to have decreased substantially in the latest period, for countries the range has in fact increased.
Looking at Figure E, one can note that unlike with country pair correlations, not all industry pair correlations increased in the pre-euro period compared to the correlations of the convergence period. Nevertheless, due to the large number of observations, the small positive average increase (0.036) is significant, so we can still state that correlations in the pre-euro period have increased. Finally, Figure F depicts that with the exception of 6 sector pairs, all correlations among industry pairs in the euro-period were lower than in the pre-euro period. Remarkably, the average decrease in this last subperiod was quite substantial (0.150 points), and significant at the 5% confidence level.
We have thus seen that both cross-country correlations and cross-industry correlations have varied very much in the different subperiods. Unfortunately, it is difficult to say whether the upswing in dispersion and the recent phenomenon of lower correlations among EMU countries is a temporary, permanent or cyclical event. Unless the recent decrease in correlations is a permanent and continuing phenomenon, the contemporary drop in itself does not bring the correlations back to the levels experienced in the era before the process of integration in Europe started (approximately 1986).
In any case, the evolution of the correlation structure over the whole sample period have lead to the fact that the difference between country correlations and industry correlations has narrowed considerably in the last subperiod. As can be seen from Table 3, country pair correlations in the last sample period (0.460) are not much lower than industry index correlations (0.541) anymore. This might indicate a possible turning point of optimal diversification strategies in the most recent time period.
Many authors have suggested the spreading of the invested position over both countries and industries as the optimal diversification strategy. If so, then the average correlation of countries with industries must be low. Columns B in Table 3 contain the average correlation of country indices with industry indices and visa versa. Average country-industry correlations are given at the last row of the table. It turns out that the average correlation of countries, from all groups and over all periods, with industries is in fact quite high: 0.525 points. Also, in all subperiods country/industry correlations exceed the level of intra-industry correlations, suggesting that diversifying over industries might be the superior asset allocation strategy, even above spreading over both countries and sectors. These results are in line with Ferreira (2003), but are in sharp contrast with Ehling and Ramos (2002) who report correlations and industries to be usually very close to zero and never above 0.2. An explanation for the difference in the results is hard to find, since the methodology of computing correlation coefficients is uniform, and the data set (including frequency) used is similar as well.
What is remarkable, is that although the level of average sector/industry correlations used to be somewhere between the average of sector correlations and country correlations, this principle seems to have ended in the last subperiod. In the euro-period, the average correlation among countries and sectors (0.571) is in this period even higher than correlations within the asset classes (0.460 and 0.541 respectively). This finding might be an indication of a shift in the asset allocation strategy by the common investor described in the literature and increasingly found in practice. When investors increasingly diversify their portfolio by looking at both countries and sectors, the correlation coefficient between the two might increase, because both assets are then equally influenced by the general market factors such as investment sentiment.
7.2.2. Discussion of results:
So far, by illustrating the time-varying pattern of correlations, we have seen that:
A. Average industry correlations in the pre-convergence period were much higher than average country pair correlations.
B. That country pair correlations have increased substantially more during the convergence and pre-euro period that industry pair correlations.
C. That correlations have decreased during the euro period, and that the average decrease for industries was higher than for countries.
D. That the trends described above are a uniform phenomenon concerning all correlation pairs, and are not simply dominated by the observations from a small number of countries/industries.
These results could indicate a possible trend in the correlation levels. Please note though that, as explained earlier, the increase in correlation between the index returns is also partially influenced by the reduced volatility and increased convergence of the exchange rates between EMU currencies over time.
The fact that correlations in the last subperiod were still much higher than the correlations in the pre-convergence period (1979 until 1986), illustrates the importance of the free market and the increased trade levels after the Single European Act. The high levels of intra-Europe trade have lead to an interdependence between European firms, possibly also leading to higher correlations of their profitabilities and stock prices. As a result, it is to be expected that correlation levels will remain at a higher level compared to the pre-convergence era, despite of the fall back in the last subperiod..
The fact that correlations have been lower in the last subperiod, is perhaps the most remarkable findings from the study of the correlations structures. The results are in contrast with the research by Goetzmann et al (2001) who find that periods of free capital flows are associated with higher correlations.
On the other hand, the results presented here are not so surprising as they do correspond to the findings of other recent studies on this topic, for example the results of Adjaoute and Danthine (2000) and Ferreira (2003) (see section 4.1). In his paper, Ferreira (2003) suggests that the low correlations in the last sub-period of the sample may be related to a reduction in the average stock return of the EMU market. However, Goetzmann et al (2001) have shown convincingly that periods of poor market performance have been associated with high correlations, rather than low correlations. In this section, I will try to give two other explanations for this remarkable development.
Referring back to the studies described in section 4.3 that describe an increasing importance of industry factors, it can be that some industry specific shock has happened in this period that has affected different countries in different ways. For example, it is well known that particularly the IT sector was severely hit by the market crash in the period after 2001. It might be that companies from the IT sector are more heavily represented in the market index of some countries (for example Nokia in Finland has a weight of 64% in the market index), as a result of which the returns of the country indices diverge more when these sectors show strong fluctuations. Of course, that would also have implied that correlations should also have been lower in the previous subperiod, when the IT sector was experiencing extraordinary positive returns. But assuming that industry effects have become more important than country effects in the recent years (as is suggested by much of the contemporary literature on this topic, see section 4.3), the rationalization given above is not implausible. As we have seen from the literature review on this topic, most studies have concluded that industry effects have indeed become more important in the recent years, perhaps even more important than country effects. If this is indeed the case, and if European countries have become increasingly specialised, then the combination of the increased importance of the sector effects and the recent collapse of the IT bubble can give a plausible explanation for the decrease in cross-country correlations.
Despite what is sometimes suggested in the literature, the fact that in the post-Euro period (when the highest level of economic and monetary integration was achieved), cross-country correlations have decreased, does not have to mean that the disappearance of currency risk and the convergence of economic policies have been unimportant for financial markets. Instead, this phenomenon might actually stress the importance of the convergence process before the introduction of the Euro.
Ever since the establishment of the Single European Act in 1986, Europe has been in a process of increased economic integration. In that transitional phase, the fiscal policies of the member states were more synchronized to a common budget norm for all member states, which put a lot of pressure on the governments of some countries to reform their economic policies. Furthermore, inflation rates were strongly reduced in all member states and interest rates and exchange rate fluctuations were diminished, all of which required deliberate action from the government of the EMU member countries. As a result, this transitional phase has lead to large identical macroeconomic shocks in most countries, which have apparently also had a strong impact on the stock markets (judging from the high correlations between national indices in that period).
However, after the introduction of the Euro and the establishment of the European Central Bank in 1999, there have been hardly any macroeconomic shocks. The fiscal policies were already synchronised to the European norms in the 1990s, and after it’s installation, the ECB has kept the inflation rates at a consistently low level, without much fluctuation among the European interest rates. In the absence of strong macroeconomic shocks, the European stock markets have possibly become less influenced by EU-wide macroeconomic shocks, and were relatively more influenced by their domestic idiosyncratic shocks.
And although financial market integration in the EMU has accelerated in the last subperiod, we have seen in a previous chapter that higher financial integration in itself does not necessarily result in higher correlations. Financial integration merely opens the channels through which international (EU-wide) shocks are transmitted, but in itself it does not amplify correlation levels. If the common risk factors do not show large fluctuations or shocks, then idiosyncratic risk factors can have a relatively larger effect on the national stock markets, leading to low cross-country correlations, even in completely integrated financial market.
It would be interesting to examine whether this decrease in correlation levels indeed unique to the euro zone, or whether this pattern can be detected worldwide since 1999. Although I have not been able to find such a study, it is not unlikely that such a global pattern is possibly occurring.
7.3 Efficient frontiers:
Figure 5A till 5D depict the mean-variance frontier of the total sample for three types of investments (diversifying over country indices only, over industry indices and over both types of indices) over the four different subperiods under consideration, as well as the whole sample period. The mean-variance frontiers are calculated assuming constant expected returns (the mean annual return from 1979-2004 of each asset class), since investors are not expected to have perfect foresight to predict the equity risk premium during each subperiod. Consequently, the visible shift in efficient frontiers of a certain asset class over the different subperiods are solely the result of the changing covariance matrices over time.
The first thing that becomes visible from looking at figures A till D is that the efficient frontiers and their relative position with respect to other investment opportunities differ greatly over the subperiods. In the first and second subsample, it seems as if diversifying over industries offers the best diversification opportunity. However, it is hard to draw firm conclusions from the graphs only, since the graphs of countries and sectors intersect in the first and the third subperiod, and so it depends on the exact position on the horizontal scale which alternative offers the best returns.
The visual assessment also strongly suggests that diversifying over both industries and countries offers the best benefits, since the country-and-sectors line is always above the two other lines in all subperiods alike. This has a very intuitively appealing rationalization: having the opportunity to invest in both country and sector indices offers the best diversification benefits, since the scope of investment alternatives has increased and so the investor can gain the highest return for a tolerated level of risk.
Although the graphs are good to visualise the opportunity set, they do not allow us to draw any firm statistical conclusions. For a more statistical assessment, we must examine Tables 5 and 6. Table 5 reports the Sharpe ratios of the tangency portfolios for each of the investment possibilities over each of the subperiods. It also reports the difference between the optimal Sharpe ratios of the country and industry portfolios, and tests whether one is statistically superior to the other.
The table reports that in subperiod 1 (Jan 1979-Dec 1985), an asset allocation strategy over countries resulted in a higher Sharpe ratio (0.694) than for sectors (Sharpe=0.585). This finding is not so surprising, considering the fact that average correlation in the pre-convergence period was much lower among the country indices than among industry indices (0.11 versus 0.38). As already mentioned, a low correlation coefficient gives better diversification opportunities, and allows a more efficient reduction of the amount of idiosyncratic risk in the portfolio. Nevertheless, the difference between the Sharpe ratios is not found to be significant.
On the other hand, during the convergence period and the pre-euro period, spreading your portfolio over industry sectors would have given the best tangency portfolio in terms of standardised return. During the convergence period (Jan 1986-Dec 1993) the difference in Sharpe ratios was actually significant, while in the pre-euro era it was not. The superiority of the sector-diversification strategy can be explained by the fact that the intra-correlation between the EMU country indices had substantially risen after 1985.
Perhaps somewhat surprising, after the introduction of the Euro, the trend seems to have reversed again. In the Euro period (Jan 1999-April 2004), diversifying over countries seems to be the best strategy again, although the Sharpe ratio of countries is not significantly higher than the Sharpe ratio of sectors. Nevertheless, this finding contradicts the findings of both Ehling et al (2002) and Moerman (2003), although both in a different respect. Ehling et al (2003) find that in the Euro subperiod (Jan 1999-Dec 2001), both diversification approaches yield identical results. Moerman (2003) found that industry indices give a better performance that country indices, and since he finds that the addition of country indices is not very valuable given an efficient industry index allocated portfolio, he concludes that it is more important to invest in industry indices than investing in country indices nowadays.
The last part of Table 5 also reports the optimal Sharpe ratio’s of the tangency portfolios of each asset class for the whole sample period (Jan 1999-April 2004). Interestingly, measured over the whole sample period consisting of 1321 observations, a diversification strategy which focused solely on the various industry sectors offered the best risk-return trade-off. The Sharpe ratio of the optimal industry allocation strategy (0.805) was significantly superior to that of the country allocation strategy (0.582). However, a careful note should be made on this last result. The finding that diversifying over industries has been superior historically, is based on sample period covering weekly returns from the period 1979 till 2004. Although such a large data sample is always nice to have for statistical purposes, the relevance of the observations from the past history can very well be questioned. After all, because of all the institutional changes mentioned before, the European markets have changed dramatically in the past 25 years, and also, many industrial sectors have fundamentally altered in nature, mainly due to the evolution in (information) technology. For this reason, not much relevance should be attached to the results concerning the whole sample period.
What becomes apparent from Table 5, and what is also visualised in the various graphs of Figure 5, is that diversification among both industries and sectors seems to be the superior strategy to the other two alternatives in all subperiods. However, as explained in a previous section, adding assets to the current portfolio (eg adding industry sectors to the current country investment strategy) always leads per definition to a shift of the efficient frontiers.
We therefore have to perform a spanning and an intersection test in order to see whether the shift of the efficient frontier is indeed significant.
The intersection test tests whether the addition of a test asset class (eg industries) is lucrative when one already holds the optimal portfolio in the other asset class (eg in countries). It specifically looks at the tangency point and checks whether an investor who is currently sitting on the tangency line of one asset (the linear combination of the efficient portfolio with the highest Sharpe ratio and the risk free asset). It tests whether that particular investor can increase his wealth by moving to a tangency line with a bigger slope (so a higher expected return per measure of risk taken) by investing in both kind of assets. The null hypothesis of the intersection test implies intersection of both tangency lines, meaning that investing in both portfolios does not give significantly higher returns at the new tangency portfolio.
Table 6A reports the values of the intersection test. The table shows that the probabilities of intersection are two low to be accepted in all subsample periods except for the convergence period. This means that unlike in the other subsamples, it can not be proven that during the convergence period, an investor who held the optimal portfolio of industry sectors could significantly improve his trade-off by adding country indices to his portfolio. This finding is in line with the results reported in Table 5, where the convergence period was the only period in which one asset class had a significantly higher Sharpe ratio than the other asset class. It is also in line with the graphical illustrations from Figure 4. Although it is a little bit hard to tell from a visual inspection only, it is not hard to imagine that the slope of the tangency line of the country index could indeed substantially different from the slope of the country and industry-tangency line. These results imply that during the convergence period (January 1986-Dec 1993), for an investor interested in holding the tangency portfolio specifically, it would be better to diversify over industries, than over either countries or both asset classes.
As explained before intersection refers to a case where the two frontiers (the benchmark frontier and the frontier of the benchmark plus test assets) intersect at a single point on the efficient frontier; the tangency portfolio. We will now look at the efficient frontier as a whole to see whether the frontier shifts at any point when new assets to the investment universe are added. For this purpose, the spanning test is conducted. Spanning refers to the case where the whole mean-variance frontier generated by an expanded investment opportunity set of benchmark assets and some test assets coincides with the frontier generated by the subset of
benchmark assets only. The results from the spanning test are reported in Table 6b. We can see that because the probabilities of accepting the null hypothesis of both spanning test are too low for acceptance. This means that neither the country indices nor the industry indices span the mean-variance frontier for both types of investment categories. In other words, a mean-variance investor can always gain by adding the other type of indices to his portfolio. This conclusion is relevant to all subperiods, including the convergence period.
Although at first notice, this result might seem to be in contradiction to the results of the above mentioned intersection test (since the intersection test accepted the null hypothesis in the convergence period), it is important to realise that the spanning test is a stricter test than the intersection test. The intersection test solely looks one point on the efficient frontier, whereas the spanning test tests whether the addition of the other asset class at any point on the efficient frontier gives significant improvements. Apparantly,although the spanning of industries in the convergence period is accepted at the tangency portfolio, it is not accepted for all other portfolios.
- MAIN CONCLUSIONS:
This paper examines the effect of the process of monetary and economic integration in the EMU on the optimal diversification strategies for the mean-variance optimising investor.
The literature review finds that most of the studies in the past focus on the relative importance of country versus sector effects in explaining correlation structures. The results from studies using this methodology are mixed and indefinite. Although there is a growing consensus that industry effects are gaining in relevance, the views diverge on which factor is in the end most important. Furthermore, although these studies do provide more insight into the underlying causes affecting cross-market correlations, their relevance for an investor who wants to determine the optimal investment strategy based on the correlation structure at hand is quite limited. The main practical lesson that investors can draw from these studies is that one should be careful not to invest in a similar industry in different countries when they diversify geographically over national indices.
An alternative research suggested by Ehling and Ramos (2002) and Moerman (2003) takes a more practical approach to the question at hand and aims to directly investigate the evolution of the optimal diversification strategy since the establishment of the EMU. Unfortunately though, the conclusions from the two studies that take this practical approach are not completely concurrent either.
This paper adds the current body of knowledge and to reconcile the findings of both studies by a) matching the data sources utilised by both studies, and b) investigating an extended sample period. The methodology of this paper follows the model specification of the studies mentioned above and is based on the portfolio theory first introduced by Markowitz (1952), as well as the mean-variance spanning test initiated by Hubermann and Kandel (1987).
The first main empirical finding reported in this paper is that correlations among country indexes have increased considerably during the whole sample period (a sign of the time-varying nature of the correlations). Although correlations among industry sectors have increased as well, the upsurge has been much larger for countries (0.47 points over the whole sample period). Remarkable is the fact that in the euro subperiod, the trend seems to have been broken, since the average correlations for both industries have fallen (though for industries more so than for countries.)
Although the driving factors behind the increase in correlation levels from 1979 to 1998 has not been explicitly investigated by this study empirically, literature and common sense suggest that it is reasonable to assume that the rationalisation can be found in the process of monetary integration within the EMU area.
In that transitional phase until 1998, the fiscal policies of the member states were more synchronized to a common budget norm for all member states, inflation rates were strongly reduced and exchange rate fluctuations were diminished, all of which required deliberate action from the government of the EMU member countries. As a result, this transitional phase has lead to large identical macroeconomic shocks in most countries, which have apparently also had a strong impact on the stock markets.
The lower correlations witnessed in the period after the introduction of the Euro are not contradictory to these argumentt. This paper argues that in the last subperiod, economic fundamentals (inflation rates, interest rates, exchange rates) and policies have converged to a stabilised low level, without large interference from government or central bank policy. As a result, although financial markets are now prone to identical macroecomic forces, in the absense of strong macroeconomic showcks, the relative importance of EU-wide risk factors has diminished compared to the idiosyncratic risk factors. Furthermore, it is argued that although financial integration has developed rapidly in the past few years, financial integration in itself does not necessarily have to lead to higher correlations per se.
A second explanation for the phenomenon of decreased correlations in the latest subperiod, can be the increased relevance importance of industry factors relative to country factors (which has been demonstrated in other studies), and the bursting of the IT-sector bubble. Assuming that the IT sector is not represented in the national indices of all countries with the same weights, and assuming that industry effects have become more important after 1999 in explaining stock returns, it is plausible that this has lead to lower correlation among country indices after the introduction of the Euro.
Turning to testing the main research question of this study, the paper both visually and statistically examines the diversification benefits of three types of investments; country indices only, industry indices only and both types of indices. As expected, the optimal tangency portfolio of the country allocation strategy outperformed the industry allocation strategy in the pre-convergence period (Jan 1979-Dec1985). Somewhat unexpectedly however, the paper finds that during the convergence period (Jan 1986-Dec 1993) industry allocation strategies outperformed diversification over countries alone. And in the pre-euro period, both strategies statistically had the same performance in terms of standardised return of their optimal tangency portfolio. In contrast to expectations raised from other literature, a country allocation strategy is still an equally performing allocation scheme as diversification over industries after the introduction of the Euro (sample period Jan 1999-April 2004).
To test whether additional diversification benefits were to be gained by diversifying over both industries and countries, intersection tests and spanning tests were conducted. The results of the intersection tests showed that during the convergence period, there were no significant diversification benefits for an investor who held the tangency portfolio of industries, to switch to the tangency portfolio of a industries-and-country allocation scheme. This is in line with the high Sharpe ratio of the industry tangency portfolio in that subperiod, which is found to be significantly higher than the Sharpe ratio of country portfolios. However, the more stringent spanning test, rejects spanning in all subsamples, including the convergence period. This leads to the conclusion that generally speaking, a diversification strategy over both industries and countries has always lead to the best risk-return trade-off.
Referring back to the studies of Ehling et al (2002) and Moerman(2003), it must be noted that unfortunately, this investigation has not been able to support the findings of neither of the papers.
Unlike Ehling and Ramos (2002), this study does not find a clear indication that country diversification strategies, after diversifying over both assets, have always been superior to industry diversification strategies before the start of the euro-era. Instead, from the beginning of 1986 till the end of 1993, industry allocation strategies would have resulted in a superior performance. And in contrast to the findings of both Moerman (2003) and Ehling and Ramos, this research cannot find that after the introduction of the Euro, diversifying over industries significantly outperforms a country asset allocation scheme.
An explanation for the difference of results can be found in the sample data employed. As explained before, Moerman makes use of MSCI indices which are based on high-capitalisation companies of the markets. This study however uses DataStream indices are defined as value-weighted broad indices of national stock markets, covering also medium- and small-capitalisation companies, correcting for the possible bias of large companies being more affected by the liberation of the European markets. And although Ehling and Ramos have used the same datasource as this study, they base their conclusions on only two years of observations after the introduction of the Euro. Fortunately, this study has the privilege of being able to use more than 5 years of post-Euro data.
All in all, the results of this paper make one question the traditional top-down allocation strategy. This approach was until recently widely used by most investment institutions and traditionally focuses on geographical diversification in the first place. More recently, the argument is made that after the introduction of the Euro, the country orientation of the top-down approach should give way, within the euro-area at least, to an industry or sector orientation.
The traditional top-down allocation strategy and the recent shift in paradigm may be questioned, because after all, the paper finds that different diversification strategies have been optimal in the four different subsample periods specified, and no clear trend is to be detected. Furthermore, this study has been unable to prove statistically that an industry allocation strategy has been superior after the introduction of the Euro (on the contrary!), so the current trend among institutional investors to restructure their asset allocation grid is not justified from pure economic/statistical perspective. Moreover, this paper clearly illustrates that diversifying over both countries and industries was the superior diversification strategy in all of the sub-periods.
Nonetheless, most institutional investors in practice have always worked according to a top-down allocation grid, and even the recent calls for reform simply suggest replacing geographical distribution by industry diversification, instead of considering both opportunities simultaneously. According to Adjaoute et al (2003), costs are a major factor why the optimal two-step allocation strategy has not been implemented in the past. Two-step allocation is costlier than a one-step strategy, and so small players could possibly only afford one step. However, as recognised by the authors themselves, although costs might be a relevant factor in the contests of doing active portfolio management, they are hard to rationalise in the context of passive strategies.
If institutional investors for any indeterminate reason wish to continue to apply a top-down allocation strategy, it is hard to make a recommendation for the optimal diversification strategy since this has seen to be subject to quite some change over time. Investors should note that there is no guarantee that diversifying over industries will permanently be the optimal investment strategy for the remaining future. As we have seen, covariances and correlations are rather volatile and indeed time-varying. This calls for a periodic revision of this research in the future for practitioners.
9. REFERENCES:
- Adam, K., T. Jappelli, A. Menichini, M. Padula and M. Pagano, ‘Study to analyze, compare and apply alternative indicators and monitoring methodologies to measure the evolution of capital market integration in the European Union’, 2002, Commission of the European Communities, Internal Market Directorate General
-
Adjaouté,K. and J.P. Danthine, 2001a, EMU and Portfolio Diversification Opportunities,Discussion paper series No. 2962, Centre for Economic Policy Research
-
Adjaouté,K. and J.P. Danthine, 2001b, Portfolio Diversification: Alive and Well in Euroland!, Discussion paper series No. 3086, Centre for Economic Policy Research
-
Adjaouté,K. and J.P. Danthine, 2002, European Financial Integration and Equity Returns: A Theory-based Assessment, Conference Paper of the Second ECB Central Banking Conference.
- Aked, M., Brightman and C., Cavaglia, S., 2000, Increasing Importance of Industry Factors, Financial Analysts Journal, vol. 56, no. 5: 41–45.
- Beckers, S., Grinold, R., Rudd, A., Stefek, D., The relative importanceof common factors across the European equity markets, 1992, Journal of Banking and Finance 16, pp. 95-96
- Brooks, R. and M. Del Negro, 2002a, The Rise in Comovement across National Stock Markets: Market Integration or IT Bubble?, Federal Reserve Bank of Atlanta, Working Paper 2002-17a
-
Brooks, R. and M. Del Negro, 2002b, International Stock Returns and Market Integration: A Regional Perspective, Federal Reserve Bank of Atlanta, Working Paper 2002-20
-
Brooks, R. and M. Del Negro, 2002c, International Diversification Strategies, Federal Reserve Bank of Atlanta, Working Paper 2002-23
- Cavaglia S., Brightman C, and M. Aked (2000), The increasing importance of industry factors, Financial Analysts Journal 56, 41-54.
-
Cooper, L. and E. Kaplanis, 1994, Home bias in equity portfolios, inflation hedging, and international capital market equilibrium, Review of Financial Studies, Vol. 7, pp. 45-60
- De Santis, G., B. Gerard, and P.Hillion (1999), The European single currency and world equity markets , in P. Hillion et al. (eds), European Capital Markets with a Single Currency, Oxford University Press
-
DeRoon, F.A. and T.E. Nijman, 2001, Testing for mean-variance spanning: a survey, Journal of Empirical Finance, Vol 8, pp.111-155
- Ehling, P. and S. Ramos, 2002, The EMU and strategies of asset alloction, University of Lausanne.
- Ferreira, MA, 2003, The importance of Industry and Country Effects in the EMU equity markets, SSRN Working Papers
-
Fratzscher, M., 2002, Financial Market Integration in Europe: On the Effects of EMU on Stock Markets, International Journal of Finance and Economics, Vol. 7 (3), pp. 165-193
- French, K. and J. Poterba, 1991, Investor Diversification and International Equity Markets, American Economic Review 81, 222-226.
- Gerard, B, P. Hillion and F. de Roon (2002), International portfolio diversification: industry, country and currency effects revisited, working paper
-
Gerard, B., P.Hillion and F.A. de Roon, 2002, International Portfolio Diversification: Industrial Structure, Country and Currency Effects Revisited, SSRN Working Paper
- Goetzmann, W., L. Li and G. Rouwenhorst, 2001, Long-Term Global Market correlations, NBER Working Paper 8612.
- Griffin, J.M. and G.A. Karolyi, 1998, Another look at the role of the industrial structure of markets for international diversification strategies, Journal of Financial Economics, Vol. 50, pp.351-373
-
Hardouvelis, G.A., D. Malliaropulos and R. Priestley, 1999, EMU and European Stock Market Integration, Discussion paper series No. 2124, Centre for Economic Policy Research
- Heckman, L., S. Narayanan and S. Patel, 1998. Country and industry importance in European returns, Salomon, Smith Barney, Working Paper.
-
Heston, S.L. and K.G. Rouwenhorst, 1994, Does industrial structure explain the benefits of international diversification?, Journal of Financial Economics, Vol. 36, pp. 3-27
-
Heston, S.L. and K.G. Rouwenhorst, 1995, Industry and Country Effects in International Stock Returns, The Journal of Portfolio Management, Spring 1995, pp. 53-58
- Huberman, G. and Kandel, 1987, Mean-Variance Spanning, Journal of Finance, 42, 873-888.
-
Isakov, D. and F. Sonney, 2002, Are practitioners right? On the relative importance of industrial factors in international stock returns, SSRN working paper, HEC-University of Geneva no 2003.04 and FAME Research Paper no 72.
- Khan, R. and G. Zhou, 2001, Mean Variance Spanning Tests, Working paper.
- Lessard, D., 1974, World, National and Industry Factors in Equity Returns, Journal of Finance 29, 379-391.
-
Lewis, K.K., 1999, Trying to Explain Home Bias in Equities and Consumption, Journal of Economic Literature, Vol. 37 (2), pp. 571-608
-
Longin, F. and B. Solnik, 1995, Is the correlation in international equity returns constant: 1960-1990?, Journal of International Money and Finance, Vol. 14 (1) pp.3-26
- Longin, F. and Solnik, B. 2001, Extreme Correlation of International Equity Markets, Journal of Finance 56(2), 649-675.
- Markowitz, H., 1952 Portfolio Selection, Journal of Finance 5, pp. 77-91
- Markowitz, H., 1959 Portfolio Selection: Efficient Diversification of Investment, Yale University Press, New Haven, CT.
- Moerman, G., 2003, ECB Working Paper Series No.327
- Obstfeld, Maurice and Alan M. Taylor, 2001, Globalization and Capital Markets, NBER Working Paper.
-
Rouwenhorst, K.G., 1999, European Equity Markets and the EMU, Financial Analyst Journal, Vol 55 (3), pp. 57-64
- Sällström, T., International Diversification Benefits; A Mean-Variance Spanning Test, 1999, Working Paper, Stockholm School of Economics
- Sharpe, W., Mutual fund performance, January 1966, Journal of Business, pp. 119-138
- Sharpe, W., The Sharpe ratio, fall 1994, Journal of Portfolio management
- Solnik, B., 1974, Why not diversify internationally rather than domestically?, 1999, Financial analysts Journal 30, 48-54
- Stulz, R. (1999), Globalization of Equity Markets and the Cost of Capital, Journal of
Applied Corporate Finance, Fall (1999), 8-25.
- Teaser, L. and I. Werner, 1995, Home Bias and High Turnover, Journal of International Money and Finance 14, 467-493.
- Treynor, J.L., and Black, F., How to use security analysis to improve portfolio selection, January 1973, Journal of Business, pp. 66-85
- Warnock, F., 2001, Home Bias and High Turnover Reconsidered, Journal of International Money and Finance 21, 795-805.
- Yang, J., Min, I., Li, Q., European stock market integration: does EMU matter?, 2003, Journal of Business Finance and Accounting, pp. 1253-1276
Depends on whether spanning test is significant or not!!