Advanced search
Add to list

The comovement of asset returns

(2006)
Author
Promoter
Jan Annaert and (UGent)
Organization
Abstract
As the saying goes: "Never put all your eggs in one basket". This old adage has become one of the cornerstones of modern portfolio theory, where it is expressed by one single word: Diversify! Indeed, investors can greatly reduce the riskiness of their portfolios by diversifying their holdings across poorly correlated assets. As a consequence, the decision on whether to add a particular asset to a portfolio will not only depend on the asset's expected return and volatility, but also on how it is correlated with the other assets in the portfolio. This implies that determining the level of asset correlations constitutes an essential component of any asset allocation strategy. For example, how much of a portfolio is invested in stocks, bonds, and cash, will depend on how these three assets are correlated. Moreover, diversification is also important when one wants to construct portfolios consisting of only one type of asset, for instance stocks. The risk of a stock market portfolio can be highly reduced by diversifying over different countries and industries. The degree of risk reduction will in this case highly depend upon the correlations between international stock markets and global industry portfolios. To have an optimal portfolio at each point in time, it is not only important to have an insight into the average level of the correlations, i.e. the unconditional correlations, but in particular into their dynamics, i.e. the conditional correlations. The literature has shown that equity return correlations vary considerably through time. For instance, cross-country equity market correlations tend to increase substantially in highly volatile bear markets or during financial crises. Correlations also vary with structural changes in the economic and financial environment. A vast literature has shown that equity market correlations tend to increase with market development and integration. Similar evidence of time-varying correlations is found across asset markets. Some studies show that stock and bond returns exhibit asymmetry in their conditional correlations. Furthermore, while stock and bond returns (within a particular market) typically exhibit a modest positive correlation in the long term, there is substantial time-variation in the relation between stock and bond returns in the short term, including sustained periods of negative correlations. For investors, it is extremely important to be able to measure and interpret the time-variation in the asset return correlations, as it may have important consequences for the way (institutional) investors should construct optimal portfolios or perform risk management. I give some examples. First, the traditionally low correlation between stock returns across countries induced investors to diversify their stock portfolios primarily on a geographical basis. Recent evidence, however, suggests that further globalization and (regional) integration have led to a significant increase in cross-country stock market correlations. An important question is hence whether geographical diversification is still the best way to reduce the total risk of international portfolios, and whether other strategies, like industry diversification, may lead to superior results. Second, there is considerable evidence that international stock market correlations are asymmetric, i.e. correlations are substantially higher in (volatile) bear markets than during bull markets. If diversification benefits from international investing are not forthcoming at the time that investors need them the most (when their home market experiences a downturn), the strong case for international investing may have to be re-considered. Third, investors may want to exploit the negative correlations between stock and bond returns typically observed during periods of increased market uncertainty. Clearly, optimal portfolios based on unconditional correlations will -- especially in these periods of high uncertainty -- be very different from those based on conditional correlations. The aim of this dissertation is threefold. First, it studies the time-variation in (ex ante) conditional correlations between international stock markets, between global industry portfolios, and between stock and bond markets. Second, it attempts to explain why correlations vary through time by relating this time-variation to economic state variables. In other words, in this dissertation, I will search for the economic fundamental sources driving the conditional correlations. The `search for fundamentals' makes this study radically different from most other studies on time-varying correlations, which were mainly concerned with modeling correlations in a purely statistical way. Moreover, while some studies have tried to explain correlations ex post, this dissertation looks essentially at the economic drivers of ex ante correlations. Obviously, it is conceivable that economic variables will not explain all of the time-variation in conditional correlations. So third, this dissertation further explores residual correlations, i.e. correlations between asset returns not explained by economic fundamentals. This analysis reveals possible non-fundamental sources of the conditional correlations. In the context of international stock markets, excess non-fundamental correlations could refer to contagion, i.e. increased correlations between stock markets in times of crisis. In the context of correlations between the stock and bond markets, non-fundamental correlations could refer to the well-known `flight-to-safety' phenomenon, where increased stock market uncertainty induces investors to flee stocks in favor of bonds, clearly generating negative correlations. Overall, the results of this dissertation will reveal information about why correlations are changing through time, which should help investors to construct optimal portfolios or perform superior risk management. The first contribution of this dissertation analyzes the dynamics and determinants of the relative benefits of geographical and industry diversification over the last 30 years. While earlier research pointed towards a clear outperformance of country over industry diversification, a number of recent papers have claimed that industry diversification benefits now outweigh those from geographical diversification. The question this paper wants to answer is to what extent the sudden relative increase in the potential of industry diversification is permanent, or whether it is merely a temporary phenomenon. Or stated in more practical terms: Should investors continue to diversify primarily across countries, or should they increasingly focus on cross-industry diversification. Time-varying market integration and development arise as key candidates to explain a reduction in the potential of geographical diversification. As said before, there is ample evidence that local stock market returns become systematically more correlated with the world (regional) stock market when the local markets become more economically and financially integrated with the world (regional) market. Unfortunately, the methodology typically used in the existing literature on country versus industry diversification is not very well suited for dealing with such structural changes. This has two implications. First, existing measures of the relative potential of geographical and industry diversification are likely to be mismeasured. Second, these studies do not yield insights into the (structural) drivers of cross-country and cross-industry correlation and risk. This contribution addresses this issue and tries to determine whether changes in international stock market correlations are due to fundamental economic sources or to non-fundamental or temporary sources. First, a new structural regime-switching volatility spillover model is developed to decompose total risk into a systematic and a country (industry) specific component. Contrary to most other studies, market betas and asset-specific risks are explicitly allowed to vary with both structural and temporary changes in the economic and financial environment. In a second step, the relative benefits of geographical and industry diversification are investigated by comparing average asset-specific volatilities and model-implied correlations across countries and industries. A large positive (negative) effect is found of the structural factors with respect to the country betas (country-specific volatility), especially in Europe, while industry betas are mainly determined by temporary factors. Not taking into account the time-variation in betas leads to biases in measures of industry and country-specific risk of up to one third. After correcting for this bias, it is found that under the influence of globalization and regional integration, the traditional dominance of geographical over industry diversification has eroded, and that over the last years geographical and industry diversification roughly yield the same diversification benefits. Finally, the results indicate that the surge in industry risk at the end of the 1990s was partly (but not fully) related to the TMT-bubble (Technology, Media and Telecom). The second contribution investigates whether European stock markets are vulnerable to contagion. In accordance with the literature we define contagion as "correlation over and above what one would expect from economic fundamentals". More specific, we use a two-factor asset pricing specification to model fundamentally-driven linkages between markets. Conditional upon the factor model, contagion is defined as correlation among the model residuals, and a corresponding test procedure is developed. Any test for contagion will, however, depend upon a correct characterization of the fundamental linkages between markets, i.e. the choice of the factors and the specification of the factor exposures. This chapter investigates to what extent conclusions from this contagion test depend upon the specification of the time-varying factor exposures. A two-factor model with global and regional market shocks as factors is developed. The global and regional market exposures are made conditional upon both a latent regime variable and two structural instruments. For a set of 14 European countries, the results show that this model outperforms more restricted versions. Using the optimal model which accounts for fundamentals, no evidence is found in favour of contagion. Opposite conclusions are obtained when more restricted versions of the factor specifications are used. The main conclusion from this chapter is that any contagion test should take into account time-varying fundamental linkages between markets. The third contribution studies the economic sources of the US stock-bond return comovement and its time-variation using a dynamic factor model. The paper starts by developing a flexible statistical model for conditional stock and bond return volatilities and correlations. An innovative feature of this model is that we use ex-post volatility and correlation measures to identify the ex ante stock-bond return correlation. We confirm previous studies that have found substantial variation in stock-bond correlations, including those episodes of (extreme) negative correlation. In the literature, a number of models have had a modest degree of success in explaining the average correlation using economic state variables. However, none of these studies try to explain the substantial time-variation in stock-bond returns correlations. The last chapter of this dissertation will not only attempt to explain the unconditional but also the time-variation in stock-bond correlations. We develop a dynamic factor model that relates both stock and bond returns to shocks in a number of economic state variables. The `fundamental' stock-bond return correlation implied by our model is the result from a joint (and possibly time-varying) exposure to these state variables as well as from their heteroskedasticity. While variables such as interest rates, inflation, output growth, and risk aversion are obvious fundamental determinants of the correlation between bond and stock returns, we also include the uncertainty about these variables as additional state variables. These variables may reflect true economic uncertainty or heteroskedasticity. Economic factors are identified employing structural and non-structural vector autoregressive models. The structural models are obtained from state-of-the-art New-Keynesian theory. The best fitting economic factor model yields a positive stock-bond return correlation that is slightly below the unconditional value observed in historical data. However, a large part of the time-variation in stock-bond return correlations remains unexplained, including the substantial negative correlation observed during the period 1999-2004. Several alternative explanations are explored for the latter phenomenon.

Citation

Please use this url to cite or link to this publication:

MLA
Inghelbrecht, Koen. The Comovement of Asset Returns. Ghent University. Faculty of Economics and Business Administration, 2006.
APA
Inghelbrecht, K. (2006). The comovement of asset returns. Ghent University. Faculty of Economics and Business Administration, Ghent, Belgium.
Chicago author-date
Inghelbrecht, Koen. 2006. “The Comovement of Asset Returns.” Ghent, Belgium: Ghent University. Faculty of Economics and Business Administration.
Chicago author-date (all authors)
Inghelbrecht, Koen. 2006. “The Comovement of Asset Returns.” Ghent, Belgium: Ghent University. Faculty of Economics and Business Administration.
Vancouver
1.
Inghelbrecht K. The comovement of asset returns. [Ghent, Belgium]: Ghent University. Faculty of Economics and Business Administration; 2006.
IEEE
[1]
K. Inghelbrecht, “The comovement of asset returns,” Ghent University. Faculty of Economics and Business Administration, Ghent, Belgium, 2006.
@phdthesis{671720,
  abstract     = {{As the saying goes: "Never put all your eggs in one basket". This old adage has become one of the cornerstones of modern portfolio theory, where it is expressed by one single word: Diversify! Indeed, investors can greatly reduce the riskiness of their portfolios by diversifying their holdings across poorly correlated assets. As a consequence, the decision on whether to add a particular asset to a portfolio will not only depend on the asset's expected return and volatility, but also on how it is correlated with the other assets in the portfolio. This implies that determining the level of asset correlations constitutes an essential component of any asset allocation strategy. For example, how much of a portfolio is invested in stocks, bonds, and cash, will depend on how these three assets are correlated. Moreover, diversification is also important when one wants to construct portfolios consisting of only one type of asset, for instance stocks. The risk of a stock market portfolio can be highly reduced by diversifying over different countries and industries. The degree of risk reduction will in this case highly depend upon the correlations between international stock markets and global industry portfolios.
To have an optimal portfolio at each point in time, it is not only important to have an insight into the average level of the correlations, i.e. the unconditional correlations, but in particular into their dynamics, i.e. the conditional correlations. The literature has shown that equity return correlations vary considerably through time. For instance, cross-country equity market correlations tend to increase substantially in highly volatile bear markets or during financial crises. Correlations also vary with structural changes in the economic and financial environment. A vast literature has shown that equity market correlations tend to increase with market development and integration. Similar evidence of time-varying correlations is found across asset markets. Some studies show that stock and bond returns exhibit asymmetry in their conditional correlations. Furthermore, while stock and bond returns (within a particular market) typically exhibit a modest positive correlation in the long term, there is substantial time-variation in the relation between stock and bond returns in the short term, including sustained periods of negative correlations.
For investors, it is extremely important to be able to measure and interpret the time-variation in the asset return correlations, as it may have important consequences for the way (institutional) investors should construct optimal portfolios or perform risk management. I give some examples. First, the traditionally low correlation between stock returns across countries induced investors to diversify their stock portfolios primarily on a geographical basis. Recent evidence, however, suggests that further globalization and (regional) integration have led to a significant increase in cross-country stock market correlations. An important question is hence whether geographical diversification is still the best way to reduce the total risk of international portfolios, and whether other strategies, like industry diversification, may lead to superior results. Second, there is considerable evidence that international stock market correlations are asymmetric, i.e. correlations are substantially higher in (volatile) bear markets than during bull markets. If diversification benefits from international investing are not forthcoming at the time that investors need them the most (when their home market experiences a downturn), the strong case for international investing may have to be re-considered. Third, investors may want to exploit the negative correlations between stock and bond returns typically observed during periods of increased market uncertainty. Clearly, optimal portfolios based on unconditional correlations will -- especially in these periods of high uncertainty -- be very different from those based on conditional correlations.
The aim of this dissertation is threefold. First, it studies the time-variation in (ex ante) conditional correlations between international stock markets, between global industry portfolios, and between stock and bond markets. Second, it attempts to explain why correlations vary through time by relating this time-variation to economic state variables. In other words, in this dissertation, I will search for the economic fundamental sources driving the conditional correlations. The `search for fundamentals' makes this study radically different from most other studies on time-varying correlations, which were mainly concerned with modeling correlations in a purely statistical way. Moreover, while some studies have tried to explain correlations ex post, this dissertation looks essentially at the economic drivers of ex ante correlations. Obviously, it is conceivable that economic variables will not explain all of the time-variation in conditional correlations. So third, this dissertation further explores residual correlations, i.e. correlations between asset returns not explained by economic fundamentals. This analysis reveals possible non-fundamental sources of the conditional correlations. In the context of international stock markets, excess non-fundamental correlations could refer to contagion, i.e. increased correlations between stock markets in times of crisis. In the context of correlations between the stock and bond markets, non-fundamental correlations could refer to the well-known `flight-to-safety' phenomenon, where increased stock market uncertainty induces investors to flee stocks in favor of bonds, clearly generating negative correlations. Overall, the results of this dissertation will reveal information about why correlations are changing through time, which should help investors to construct optimal portfolios or perform superior risk management.
The first contribution of this dissertation analyzes the dynamics and determinants of the relative benefits of geographical and industry diversification over the last 30 years. While earlier research pointed towards a clear outperformance of country over industry diversification, a number of recent papers have claimed that industry diversification benefits now outweigh those from geographical diversification. The question this paper wants to answer is to what extent the sudden relative increase in the potential of industry diversification is permanent, or whether it is merely a temporary phenomenon. Or stated in more practical terms: Should investors continue to diversify primarily across countries, or should they increasingly focus on cross-industry diversification. Time-varying market integration and development arise as key candidates to explain a reduction in the potential of geographical diversification. As said before, there is ample evidence that local stock market returns become systematically more correlated with the world (regional) stock market when the local markets become more economically and financially integrated with the world (regional) market. Unfortunately, the methodology typically used in the existing literature on country versus industry diversification is not very well suited for dealing with such structural changes. This has two implications. First, existing measures of the relative potential of geographical and industry diversification are likely to be mismeasured. Second, these studies do not yield insights into the (structural) drivers of cross-country and cross-industry correlation and risk. This contribution addresses this issue and tries to determine whether changes in international stock market correlations are due to fundamental economic sources or to non-fundamental or temporary sources. First, a new structural regime-switching volatility spillover model is developed to decompose total risk into a systematic and a country (industry) specific component. Contrary to most other studies, market betas and asset-specific risks are explicitly allowed to vary with both structural and temporary changes in the economic and financial environment. In a second step, the relative benefits of geographical and industry diversification are investigated by comparing average asset-specific volatilities and model-implied correlations across countries and industries. A large positive (negative) effect is found of the structural factors with respect to the country betas (country-specific volatility), especially in Europe, while industry betas are mainly determined by temporary factors. Not taking into account the time-variation in betas leads to biases in measures of industry and country-specific risk of up to one third. After correcting for this bias, it is found that under the influence of globalization and regional integration, the traditional dominance of geographical over industry diversification has eroded, and that over the last years geographical and industry diversification roughly yield the same diversification benefits. Finally, the results indicate that the surge in industry risk at the end of the 1990s was partly (but not fully) related to the TMT-bubble (Technology, Media and Telecom).
The second contribution investigates whether European stock markets are vulnerable to contagion. In accordance with the literature we define contagion as "correlation over and above what one would expect from economic fundamentals". More specific, we use a two-factor asset pricing specification to model fundamentally-driven linkages between markets. Conditional upon the factor model, contagion is defined as correlation among the model residuals, and a corresponding test procedure is developed. Any test for contagion will, however, depend upon a correct characterization of the fundamental linkages between markets, i.e. the choice of the factors and the specification of the factor exposures. This chapter investigates to what extent conclusions from this contagion test depend upon the specification of the time-varying factor exposures. A two-factor model with global and regional market shocks as factors is developed. The global and regional market exposures are made conditional upon both a latent regime variable and two structural instruments. For a set of 14 European countries, the results show that this model outperforms more restricted versions. Using the optimal model which accounts for fundamentals, no evidence is found in favour of contagion. Opposite conclusions are obtained when more restricted versions of the factor specifications are used. The main conclusion from this chapter is that any contagion test should take into account time-varying fundamental linkages between markets.
The third contribution studies the economic sources of the US stock-bond return comovement and its time-variation using a dynamic factor model. The paper starts by developing a flexible statistical model for conditional stock and bond return volatilities and correlations. An innovative feature of this model is that we use ex-post volatility and correlation measures to identify the ex ante stock-bond return correlation. We confirm previous studies that have found substantial variation in stock-bond correlations, including those episodes of (extreme) negative correlation. In the literature, a number of models have had a modest degree of success in explaining the average correlation using economic state variables. However, none of these studies try to explain the substantial time-variation in stock-bond returns correlations. The last chapter of this dissertation will not only attempt to explain the unconditional but also the time-variation in stock-bond correlations. We develop a dynamic factor model that relates both stock and bond returns to shocks in a number of economic state variables. The `fundamental' stock-bond return correlation implied by our model is the result from a joint (and possibly time-varying) exposure to these state variables as well as from their heteroskedasticity. While variables such as interest rates, inflation, output growth, and risk aversion are obvious fundamental determinants of the correlation between bond and stock returns, we also include the uncertainty about these variables as additional state variables. These variables may reflect true economic uncertainty or heteroskedasticity. Economic factors are identified employing structural and non-structural vector autoregressive models. The structural models are obtained from state-of-the-art New-Keynesian theory. The best fitting economic factor model yields a positive stock-bond return correlation that is slightly below the unconditional value observed in historical data. However, a large part of the time-variation in stock-bond return correlations remains unexplained, including the substantial negative correlation observed during the period 1999-2004. Several alternative explanations are explored for the latter phenomenon.}},
  author       = {{Inghelbrecht, Koen}},
  language     = {{eng}},
  pages        = {{XXII, 183}},
  publisher    = {{Ghent University. Faculty of Economics and Business Administration}},
  school       = {{Ghent University}},
  title        = {{The comovement of asset returns}},
  year         = {{2006}},
}