markets. We estimate a conditional asset pricing model, which allows for a time-varying degree of integration that measures the importance of EU-wide ...

1 downloads 4 Views 748KB Size

JEL Classi cation: G12, G15 Keywords: CAPM, Integration of Stock Markets, European Monetary Union. Hardouvelis is at the University of Piraeus, the Prime Ministers' Oce, and CEPR. Malliaropulos is at the University of Piraeus and the National Bank of Greece. Priestley is at the Norwegian School of Management. Address for correspondence: Dimitrios Malliaropulos, Strategic Planning and Research Division, National Bank of Greece, 86 Eolou Street, 10232 Athens, Greece. e-mail: [email protected] We would like to thank Bernard Dumas, Bruno Solnik, Warren Bailey, Bruno Gerard and Sergei Sarkissian, for helpful comments. The paper has bene tted from useful discussions with seminar participants at Norges Bank, HEC School of Management, Foundation Banque de France, Norwegian School of Management, Norwegian School of Economics and Business Administration, Stockholm School of Economics, University of Cyprus, the Center of Financial Studies in Frankfurt, and the Annual Meetings of the European Finance Association and the European Financial Management Association. Rene Stulz, the editor, and an anonymous referee provided detailed comments and many valuable insights which improved the article. An earlier version of the paper was circulated as a CEPR Discussion Paper No. 2124.

EMU and European Stock Market Integration

1

1 Introduction The central question in this paper is whether or not the intense eorts of the European countries in the 1990s to establish the European Monetary Union (EMU), which led to money market and bond market integration, also led to gradual stock market integration. On January 1, 1999, eleven European Union (EU) countries, namely Austria, Belgium, Finland, France, Germany, Ireland, Italy, Luxembourg, Netherlands, Portugal and Spain, formed a monetary union. Since then, the exchange rates between the EU11 countries have been irrevocably xed, the euro was introduced as the common currency, the European Central Bank (ECB) began operating, carrying out the common monetary policy, and all EU-11 government bills and bonds are denominated in euro. Money market and bond market integration was an immediate consequence of EMU. In fact, because nancial markets are forward looking and because by 1997 they were convinced that with very high probability 11 of the 15 EU countries would eventually join EMU, bond market integration had taken place among those 11 countries as early as 1997. Moreover, a country's long-term interest rate spread with the corresponding interest rates of Germany, the anchor country, served as an indicator of the probability that the country would eventually manage to join EMU.1 Stock market integration is another possible implication of EMU because the introduction of the common currency eliminates many obstacles to intra-European portfolio allocation. The common currency eliminated intra-European exchange rate uctuations and, hence, the cost of hedging that risk. It also nulli ed the importance of various restrictions on the foreign currency composition of assets held by pension funds, insurance companies and other local institutions. These institutions are typically restricted by currency matching rules or by maximum weights on assets denominated in foreign currency. The adoption of euro nulli ed such legal restrictions within EMU without a change in the domestic law. For investors with long horizons, the cost of hedging intra-European currency risk or the cost of avoiding various currency restrictions on their portfolio composition are reduced as the probability of a country joining EMU increases and the time to the launch of the single currency approaches. Home bias due to currency habitats should be reduced as the costs of hedging See, for example, JP Morgan: \The EMU Calculator", October 1996, and \EMU Calculator Handbook", January 1997; Paribas: \EMU Countdown", February 1997; Credito Italiano: \Economic Trends in Italy", IV 1996; Goldman Sachs: \European Bond Spreads and the Probability of EMU", May 1996; Favero et.al. (1997). 1

EMU and European Stock Market Integration

2

EU-11 currency risk fall and currency restrictions are nulli ed, thus leading domestic investors to increase their holdings of pan-European assets.2 One way to assess whether European stock markets have become more integrated during the 1990s is to examine whether or not the in uence of country-speci c risk factors on required stock returns decreases in favor of EU-wide factors. For this reason we estimate a conditional asset pricing model of European stock markets, which incorporates an explicit time-varying degree of integration. The time-varying degree of integration measures the importance of EU-wide relative to country-speci c risk factors. A major innovation in our model is the incorporation of an explicit link between the probability of a country joining EMU and the degree of a country's stock market integration. The probability of joining EMU is modeled as a function of a country's expected interest rate dierential vis-a-vis Germany for the year when EMU was expected to begin. This probability also captures | in an inverse manner | the expected cost of hedging intraEuropean currency risk and avoiding intra-European currency restrictions. An increase in this probability should lead to a shift in the pricing of local assets away from local factors towards EU-wide factors, that is, there should be changes in the level of integration given the expectation of a change in investment opportunities. Another attractive feature of our model is that it accounts for time-variation in both quantities and prices of risk. The latter are conditioned on EU-wide and country-speci c information variables. Overall, the model allows us to recover a conditional measure of integration of European stock markets and to quantify the relevance of local and global risk in determining expected returns over time. The rest of the paper is organized as follows: Section 2 discusses the relationship between the introduction of the single currency and stock market integration. Sections 3 and 4 present our asset pricing model and its econometric implementation. Section 5 describes the data. Section 6 contains the main empirical results along with a battery of speci cation tests. Section 7 concludes. There are two additional reasons why the integration of European stock markets may have been enhanced. First, convergence of interest rates and in ation rates towards German levels leads to convergence of real risk-free rates across EMU member states. Second, real convergence of European economies leads to convergence of expected real cash ows across markets. 2

EMU and European Stock Market Integration

3

2 The Single Currency and Stock Market Integration Integration of European stock markets is de ned both in terms of the type of risk investors are exposed to and in terms of their reward to risk relationship. Complete integration assumes investors face only common EU-wide risk and price it identically (see Solnik (1974)). Complete segmentation assumes investors face and price only country-speci c risk. Partial integration assumes that, in addition to common EU-wide risk, investors face country-speci c risk and price them both. Partial integration can arise from ocial barriers to international investment or other information and trading costs. The European experience of the 1990s ts the framework of partial integration because, although most ocial restrictions on intra-European investment ows were lifted by 1993 (Licht (1998)), up to January 1, 1999, the cost of hedging intra-European currency risk and other currency restrictions continued to act as a barrier to intra-European asset allocation, thus creating home bias. Models of partial integration assume that there exist barriers to international investment. These models are discussed in Black (1974), Stehle (1977), Stulz (1981), Errunza and Losq (1985) and Cooper and Kaplanis (1999), among others. Unlike standard models of international portfolio choice, models of partial integration predict that investors' portfolios can be biased towards home assets. Stulz (1981) constructs a model of international asset pricing in which domestic investors face a cost for holding risky foreign securities as a result of barriers to international investment, whereas foreign investors face no such cost. Both domestic and foreign investors can buy or sell a risk-free bond and hold short positions of risky assets. There are two countries with a nite number of (identical) investors and risky securities. Each investor maximizes a utility function which depends positively on end-of-period expected wealth and negatively on the variance of wealth. For a given vector of expected excess returns and covariance matrix of returns, Stulz shows that investors exclude from their portfolio those foreign assets which do not provide an expected return large enough to oset the cost of holding them. Cooper and Kaplanis (1999) expand the Stulz model in a multi-country setting and clarify further the dierences between the expected return on `local' assets, which are held only by domestic investors, and the expected return of `global' assets, which are held by both domestic and foreign investors. Local assets can be thought o as domestic assets which are excluded from foreign investors' portfolios because transaction costs are higher than the diversi cation bene t from including these assets in the global portfolio.

EMU and European Stock Market Integration

4

The expected return of a global asset is determined by a wealth-weighted cost term and the asset's covariance with the global portfolio. The expected return on a composite local asset { the domestic stock price index { is determined (in addition to a cost term) by a premium for its covariance with the global market index (market risk), and a premium for the part of its risk that is unrelated to global assets (local risk). The relative importance of local risk depends on the proportion of local assets in world market value. A similar result is also obtained in Errunza and Losq (1985), where some investors are precluded from holding some securities.3 The partial integration model described above ts the European experience in the 1990s. Although most barriers to free capital mobility were lifted in Europe by 1993, some restrictions related to the preferred currency composition of institutions such as pension funds or life insurance companies remained. Typically, in most countries, these institutions were required to hold assets primarily in domestic currency. In France, for example, pension funds must have a 95% currency match between liabilities and assets. In Germany the level is 80%. In addition, the uctuation of intra-European exchange rates imposed hedging costs for all investors | the interest rate dierentials were frequently quite large | on cross-border European asset allocation.4 Hence, during the 1990s, the pricing of European stocks ought to re ect both local and EU-wide risk. On January 1, 1999, the emergence of the euro nulli ed the restrictions on the intra-European currency composition of pension funds and life insurance companies and eliminated the costs of hedging intra-European currency uctuations. This event implies that investors should include more European stocks in their portfolios as the expected return on most stocks outweigh the trading costs. Moreover, the creation of the euro did not come as a surprise, but was widely anticipated from the early 1990s, at least since February 1992, when the Maastricht Treaty was signed. Hence, during the 1990s, the expectation of the future elimination of hedging costs and other restrictions on asset allocation ought to have aected stock prices. As the 3 In Errunza and Losq's (1985) speci cation of expected returns of local (`ineligible') assets, the relative weight of their own variance depends negatively on the squared correlation between local and global (`eligible') assets. When this correlation tends to unity, the local risk premium disappears. 4 Although average volatility of intra-European exchange rates in the 1990s was lower than volatility against the US dollar and the Japanese yen, the long swings of the Deutschmark during this period suggest that intra-European currency risk was in fact signi cant. During the period 1991-1995 the Deutschmark appreciated by more than 30% in real terms relative to EU currencies, whereas during the period 1996-1998 the Deutschmark depreciated by around 15% against the same currencies.

EMU and European Stock Market Integration

5

time to January 1, 1999 was approaching, the anticipated reduction in trading costs and other restrictions was, by construction, becoming more and more important in investment decisions. Similarly, as the probability of a country joining the euro-zone was increasing or decreasing, the anticipated importance of the future reduction in trading costs was rising or declining.5 It follows that a model of partial stock market integration that purports to explain the European experience of the 1990s ought to incorporate features associated with the likelihood and the time of EMU occurring. There is already evidence consistent with the view that the anticipated introduction of the single currency has led to a substantial increase in cross border investment ows. In the early 1990s intra-EU cross border annual equity ows were around $50-$60 billion, whereas by mid 1998 they had risen to $120-$140 billion, and it is estimated that up to $1.5 trillion of rebalancing of equity portfolios (more than one third of market capitalization) will take place in the EU countries as investors switch from domestic to pan-European portfolios (Euromoney, August 1998). A survey of European clients by Dresdner Kleinwort Benson in April-May 1998 revealed that 53% of clients' equity holdings were expected to be shifted from a domestic to a EU portfolio by 2001.6 In the UK and the Netherlands, where there are essentially no formal restrictions on foreign investment, the portfolio composition of funds is distinctly dierent from the rest of the EU. Blake et.al. (1997) report that, in 1994, UK pension funds invested 22.5% of their assets in foreign equities. In contrast, Lewis (1999) reports that the percentage of foreign assets held by German institutional investors in 1993 was 4.5%.

3 The Empirical Model The empirical model is based on the intuition of partial integration discussed in the previous section. The conditional mean excess return on the ith stock market index can be written as:

Roland (1999) shows that proportional transaction costs of foreign investments can have a large impact on the extent of home bias. 6 Similar patterns in portfolio shifts are reported in a Goldman Sachs and Watson Wyatt survey of fund managers, conducted in March 1998 | see Euromoney, August 1998. They report that by August 1998, 27% of fund managers had already implemented some change and 52% were well advanced in their preparations for EMU. Furthermore, 75% of fund managers indicated that they would be reconsidering their asset allocation as a direct result of EMU. Only 9% of pension funds said that they would continue with a country based allocation strategy. 5

EMU and European Stock Market Integration

Et;1 [ri;t ]

6

= i;t;1 (EU;t;1 covt;1 [ri;t ; rEU;t ] + C;t;1 covt;1 [ri;t ; rc;t ]) +(1 ; i;t;1 )i;t;1 vart;1 [ri;t ] (1)

where ri;t is the excess return on the local stock market index, rEU;t is the excess return on the EU stock market index, rc;t is the excess return on the currency7 , EU;t;1 is the price of EU market risk, , C;t;1 is the price of currency risk, i;t;1 is the price of local risk, covt;1 is the conditional covariance operator, vart;1 is the conditional variance operator, Et;1 is the conditional expectations operator, given information up to time t ; 1; and i;t;1 measures the conditional level of integration of market i based on information up to time t ; 1 (0 i 1).8 Equation (1) may be viewed as an approximation of expected returns in partially integrated markets where both global and local risk factors are priced and the degree of integration is evolving over time. At a given moment in time, the model resembles the static formulation in Cooper and Kaplanis (1999) or Errunza and Losq (1985), with a few dierences. First, in those models the two risk factors are orthogonal, whereas we do not impose such a restriction in our framework because we tie our factors to observable portfolios. Second, in our model, i;t;1 is proxied by a function of the forward interest rate dierential between country i and Germany because this differential represents the expected cost of hedging currency risk and theory suggests that as the expected cost of hedging declines the importance of global risk increases. Third, in Cooper and Kaplanis and in Errunza and Losq, the global risk factor is the covariance with a portfolio of `global' (`eligible') assets which are held by both domestic and foreign investors. This portfolio is unobservable. In our empirical model the global portfolio is the value-weighted EU-12 portfolio, as will be described later. Fourth, our global factor includes a currency risk component, C;t;1 covt;1 [ri;t ; rc;t ] ; due to deviations from Purchasing Power Parity, namely changes in nominal exchange rates that do not match perfectly in ation dierentials. This risk is faced by investors even in a perfectly integrated global asset market due In theory, r includes the dierential in ation as well. This component is excluded in the empirical analysis because it is very small in the weekly horizon, which we utilize in the estimation. Dumas and Solnik (1995) and other authors make a similar assumption at a monthly horizon. 8 Alternatively, in the context of a regime switching model, ;1 could be interpreted as the conditional probability that market i is fully integrated (Bekaert and Harvey (1995, 1997)). In Cooper and Kaplanis (1999), the parameter ;1 can be interpreted as the relative weight of `global' assets in total market value. 7

c;t

i;t

i;t

EMU and European Stock Market Integration

7

to dierences in purchasing power indices across countries (see Adler and Dumas (1983)). The time-varying parameter i;t;1 is modelled as follows: i;t;1 = g0;i + exp (g1;i jsi;t;1 j)

(2)

where si;t;1 is the forward interest rate dierential between country i and Germany, the benchmark country, g0;i ; g1;i are country-speci c parameters and exp(:) denotes exponentiation. When the forward interest rate dierential, si;t;1 ; is zero, exp (g1;i jsi;t;1 j) becomes unity. When si;t;1 deviates from zero and g1;i < 0, then 0 < exp (g1;i jsi;t;1 j) < 1. The larger the deviation of si;t;1 from zero, whether it is positive or negative, the closer exp (g1;i jsi;t;1 j) is to zero. The constant term g0;i acts as an intercept correction on the level of integration. For example, spreads may be zero by chance but markets may not be fully integrated due to, say, capital market or ownership restrictions. When equation (2) is estimated, g1;i is negative and g0;i is very close to zero, thus the level of integration i;t;1 ends up being bounded between zero and unity. It is important to stress that the level of the forward interest rate dierential between two countries has a dual role. First, it re ects the expected change in the exchange rate between two future points in time. After January 1, 1999 the exchange rates of the EMU countries are xed. Consequently, if market participants expect that, country i and Germany will be in EMU in 1999, then the forward interest rate dierential between country i and Germany, which connects two future points in time | both dates occurring after January 1, 1999 | ought to be zero. Conversely, if market participants expect that country i will not be in EMU in 1999, or that EMU itself may not materialize, then the forward interest rate dierential between country i and Germany can be dierent from zero. Thus, market expectations of economic and monetary integration are incorporated into forward interest rate dierentials. Second, the forward interest rate dierential represents the costs of hedging foreign exchange risk within the EMU area between two future points in time. A zero forward interest rate differential implies that costs of hedging foreign exchange risk and avoiding other currency restrictions within the EMU area are expected to disappear. Hence, with a declining forward interest dierential, the share of local assets in investors' portfolios should decrease and, consequently, the weight of EU-wide risk in required returns, i;t;1 ; should increase. A rst piece of preliminary evidence that forward interest rate dierentials vis-a-vis Germany could be related to the degree of integration of

EMU and European Stock Market Integration

8

European stock markets comes from simple graphical analysis. Figure 1 displays on the left scale a 52-week ahead rolling estimate of the average cross-correlation coecient between local stock returns and the EU-12 market return, where the latter is computed as the value-weighted return of the eleven EMU countries plus the UK. The gure also displays on the right scale the average forward interest rate dierential of the seven countries for which data on interest rate swaps are available over a longer sample.9 The average cross-correlation between local returns and EU-12 returns increases signi cantly over the sample from less than 0.5 in 1991 to around 0.75 in June 1997. Over the same period, the average forward dierential uctuates in the opposite direction and decreases from more than 170 basis points to less than 25 basis points. The sample correlation coecient between the average forward interest rate dierential and the average cross-correlation of local returns and EU-12 returns is -0.88, suggesting a strong negative association between the series.10 This correlation is similar whether or not we exclude the volatile EMS crisis period of 1992-1993.

4 Econometric Speci cation and Estimation Before estimation, it is also necessary to establish a model for the EU market return, the currency return, the second moments of returns, and the prices of risk. The full model to be estimated is described below by a system of seven equations (3)-(9) plus the earlier equation (2). The rst three equations describe the level of returns: rEU;t = EU;t;1 vart;1 [rEU;t] + C;t;1 covt;1 [rEU;t ; rC;t ] + "EU;t (3) rC;t = EU;t;1 covt;1 [rEU;t; rC;t ] + C;t;1 vart;1 [rC;t ] + "C;t (4) = i;t;1 (EU;t;1 covt;1 [ri;t ; rEU;t ] + C;t;1 covt;1 [ri;t ; rC;t ]) +(1 ; i;t;1 )i;t;1 vart;1 [ri;t ] + "i;t (5) where "t = ["EU;t ; "C;t ; "i;t ] j Xt;1 N (0; Ht ); is the vector of unexpected excess returns, given the set of information X available at time t;1; Ht is the ri;t

Belgium-Luxembourg, France, Germany, Italy, Netherlands, Spain and the UK. One could argue that the increasing correlation of the national stock markets with the EU-wide portfolio during the 1990s is originating from accidental, increasingly more symmetric, shocks hitting the EU economies. It is dicult, however, for such a hypothesis to accomodate the negative association of this correlation with the forward interest dierentials. 9

10

EMU and European Stock Market Integration

9

conditional variance-covariance matrix of excess returns, and the parameter i;t;1 is determined by equation (2). The fourth equation of the system describes the conditional variancecovariance matrix of excess returns, Ht ; which follows a GARCH(1,1) process (see Baba et.al. (1989) ):

H = C0C + A0" ;1 "0 ;1 A + B0 H ;1B (6) where for N assets C is a (N + 2 N + 2) symmetric matrix and A and B t

t

t

t

are (N + 2 N + 2) diagonal coecient matrices. The last three equations of the system, equations (7)-(9), specify the evolution of the conditional prices of risk: EU;t;1 = exp 0EU XEU t;1 C;t;1 = 0C XEU t;1

i;t;1 = exp 0i XLi;t;1

(7) (8) (9)

where XEU represents EU-wide information variables, XL represents local information variables speci c to country i and 0EU ; 0C and 0i are vectors of coecients. The functional form of EU;t;1 ; C;t;1 and i;t;1 is dictated by the implications of the theoretical model. Under risk aversion, the prices of risk EU;t;1 and i;t;1 must be always positive (see Merton (1980) and Adler and Dumas (1983)). Therefore we assume that EU;t;1 and i;t;1 are exponential functions of the instruments. However, the theory does not impose any restrictions on the sign of the price of currency risk, since market participants may be willing to attach a negative price to currency deposits if their expected return in excess of the risk-free rate is negative and currency returns co-vary positively with the market. Therefore, a linear speci cation is chosen for C;t;1 : The parameters are estimated rst using the SIMPLEX algorithm based on starting values obtained from OLS and NLLS and then by maximum likelihood (ML) assuming conditional normally distributed errors. In order to avoid problems due to non-normality in excess returns we provide QuasiML (QML) estimates, as proposed by Bollerslev and Wooldridge (1992), which are robust to departures from normality. Given the highly nonlinear structure of the model and the large number of parameters involved in estimation, we estimate the model for each country in two steps. First we estimate a bivariate model of the EU market returns

EMU and European Stock Market Integration

10

and currency returns | equations (3) and (4) plus the relevant variancecovariance elements of equation (6). This provides estimates of the price of EU market risk, the price of currency risk, the conditional variances and the covariance of the EU market excess return and the excess currency return. In order to maintain the assumption that these prices of risk are equal across countries, we then impose these estimates on a set of N bivariate equations, one for each country along with the EU index and the excess currency return. This strategy necessarily leads to some loss of eciency. However, a simultaneous estimation of the full model is not practically feasible, given that | with 11 national markets, ve local and ve global instruments | a total of 124 parameters would need to be estimated simultaneously.

5 Data and Preliminary Analysis We use weekly, Deutschmark-denominated, total (i.e. dividend adjusted) continuously compounded stock returns based on Friday closing prices on the eleven EU countries. We also include the United Kingdom, which has yet to decide when and if it will enter EMU, because of its large market capitalization. Belgium-Luxembourg are aggregated into one single market. For simplicity, we treat Europe as the universe of investment opportunities available to a European investor. Returns on the EU-12 benchmark index are based on a capitalization-weighted equity price index of the EMU-11 countries plus the UK. The data source is Datastream International. The sample begins in January 1991 and ends in June 1998 (390 weekly observations).11 The weekly excess currency return is calculated as the continuously compounded dierence in the Eurocurrency interest rates between a given country and Germany, adjusted for the rate of depreciation vis-a-vis the Deutschmark.12 Hence, currency risk is measured by the ex post deviation from uncovered interest parity vis-a-vis the Deutschmark. Eurocurrency interest rates are London Friday closing rates and Deutschmark exchange rates are Frankfurt stock exchange xings from Datastream. Also, the yields employed come from the relatively more liquid one-month euro-deposit market. The use of euro-deposits with maturity of one month instead of one week | the latter are not easily available | does not create any problems The beginning of the sample is constrained by the availability of data on interest rate swaps, which are used in the construction of forward interest rate dierentials. 12 The weekly excess return for each individual currency is computed as: r ln(e ) ; ln(e ;1 ) + [ 521 ln(1 + R ;1 ) ; 521 ln(1 + R ;1 )], where e is the exchange rate (Deutschmark per unit of currency i) and R , R are the annualized one-month eurocurrency interest rates of currency i and the Deutschmark, respectively. 11

c;i;t

i;t

i;t

i;t

GE;t

i

GE

i;t

EMU and European Stock Market Integration

11

if the term structure of interest rates of country i in the horizon between one week and one month is the same as that of Germany. The currency basket is the trade-weighted excess currency return vis-avis the Deutschmark of the six most actively traded European currencies, the British pound, the French franc, the Italian lira, the Belgian franc, the Dutch guilder, and the Spanish peseta. The weights used to construct the aggregate measure of currency risk are the 1994 export shares of each country in total intra-EC-trade, adjusted such that they add up to unity.13 To model expected returns we choose a set of instruments that have been shown to be useful in predicting returns.14 For the EU market we chose a constant, the rst lag of the EU-12 index dividend yield in excess of the one-month, annualized, euro-DM deposit rate (DYIR), the rst lag of the change in the term structure (T S ), the rst lag of the change in the one-month ECU deposit rate (SIR), and the rst lag of the default spread (DS ).15 The local instruments are a constant, the rst lag of the local market index dividend yield in excess of the local market one-month deposit rate, the rst lag of the change in the local short term interest rate, the rst lag of the change in the local market term structure (de ned as the spread between the yield on ten-year benchmark government bonds and the one-month euro-deposit rate of the local currency), and the rst lag of the local market excess return.16 All data are taken from Datastream International.

5.1 Forward Interest Rate Dierentials

For each of the EU-12 countries, we obtain forward interest rate dierentials which are calculated from swap rates between xed and oating rate govThe trade-weighted aggregate excess currency return is computed as: r (0:06r +0:106r +0:061r +0:057r +0:038r +0:077r )=0:399, where 0.399 is the sum of the six individual country export shares in total intra-EC trade. The source for trade weights is \European Economy", European Commission, DG II, No. 64, 1997, Table 45, column EUR12. 14 See, for example, Campbell (1987), Harvey (1991), Ferson and Harvey (1993, 1994), Bekaert and Harvey (1995), Hardouvelis et.al. (1996) and De Santis and Gerard (1997). 15 T S is de ned as the change in the spread between the yield on ECU bonds with ten years to maturity and the one-month ECU deposit rate. DS is de ned as the spread between a weighted average of the corporate bond yields in France, Germany, Italy, Netherlands, Spain and the UK (weights based on stock market capitalizations) and the yield on the ten-year ECU bond. Corporate bond yields are from \The Economist". 16 For Ireland, where no short term interest rate is available we replace the change in the short term rate with the change in the long term rate and omit the term structure instrument. 13

c;t

c;BL;t

c;F R;t

c;I T ;t

c;N L;t

c;SP;t

c;U K;t

EMU and European Stock Market Integration

12

ernment bonds as follows: De ne wi;;t as the swap rate at time t in country i for an interest rate contract in which the interest payments of a variable rate government bond with years to maturity are exchanged against the interest payments of a xed rate government bond with the same years to maturity and the same notional principal on which the interest payments n are based. Let fi;T;t denote the n-year forward rate T years from now for country i. From the swap rates we can calculate the forward rates as: "

+ wi;T +n;t )T +n n fi;T;t = (1 (1 + wi;T;t )T

#1

n

;1

We set n = 8 and T = 2 and hence calculate for each market the eight year forward rate in two years' time. Subsequently, we calculate spreads for each market vis-a-vis Germany, the anchor-country, as: si;t = ln(1 + 8 fi;82;t ) ; ln(1 + fGE; 2;t): For Germany itself, we construct the spread between the German forward rate and the ECU forward rate. These forward interest rate dierentials provide a measure of convergence towards EMU, which is independent of the stock market. They have been widely used by market participants in order to assess the probability of individual countries to participate in EMU (see references in footnote 1). Although over longer periods of time forward interest dierentials ought to be stationary since they correspond to future forward premia in the foreign exchange market, nonstationarity may be a problem in small samples.17 Since European interest rates were on a downward trend since 1992, forward interest dierentials may have co-trended with interest rates. In this case it would be more appropriate to measure forward interest dierentials relative to the level of interest rates. Therefore, in our empirical analysis we use the ratio between the forward interest dierential and the German long bond yield. Interest rate swap yields are weekly (Friday) quotes collected from Datastream International. Datastream quotes the all-in cost of the xed-side of the swap contract for maturities from one year to ten years. The swap rate represents the interest rate paid by the xed-rate party for receiving the variable interest rate. For most of the countries the sample covers the period 29:6:91 to 26:6:98 except for Portugal and Austria (6:1:95 to 26:6:98), Finland (18:10:96 to 26:6:98) and Ireland (2:8:96 to 26:6:98). In calculating In fact, standard Dickey Fuller tests for a unit root cannot reject the null hypothesis that forward interest rate dierentials are nonstationary in ve out of eleven cases: Germany, France, Italy, Spain, Portugal. 17

EMU and European Stock Market Integration

13

interest rate swap yields, we adjust for market conventions in national swap markets.18

6 Empirical Results Before we consider each market separately, the EU-12 market and currency expected return and (co-)variance must be estimated. The rst two rows of Panel A of Table I reports results from estimating the price of EU-12 market risk and the price of currency risk associated with uctuations of a tradeweighted basket of European currencies vis-a-vis the Deutschmark. The estimated model is equations (3), (4), (6), (7) and (8). All the coecients in the conditional (co-)variance equations (not reported) and most of the coecients on the EU instruments are signi cantly dierent from zero at the 1% con dence level. This suggests time-variation in both prices and quantities of risk. The remaining rows of Panel A report individual country estimates which are obtained by augmenting equations (3), (4), (6), (7) and (8) with equations (2), (5) and (9) and reestimating whilst imposing the estimates from (3), (4), (6), (7) and (8). Many of the local instruments are signi cant in predicting the local price of risk, suggesting time variation. The estimate of g1 is negative, as expected, and signi cantly dierent from zero in all cases, suggesting that the degree of integration with EU-12 increases as forward spreads approach zero and the probability of joining EMU increases. Figure 2 displays the estimated integration weight for each local market. There is a clear upward trend in the integration weight for every local market. All the weights approach unity at the end of the sample with the interesting exception of the UK which has not signed up to the single currency. There are similar cross-country patterns in the integration weights. After the ERM crisis in September 1992 there was a general increase in the degree of integration as the Bundesbank began a policy of gradual monetary easing and ERM uctuation bands were widened to 15%, allowing interest rate dierentials vis-a-vis Germany to decline and increasing the likelihood of a future monetary union. The decrease in the degree of integration during the rst half of 1994 coincides with the international bond market crisis, triggered by step-wise increases in US short-term interest rates. Concerns Coupon payments are usually made annually, but in Ireland and the UK they are made semi-annually. We annualize interest rate swap yields in these countries accordingly in order to derive comparable returns with Germany. Also, in Belgium, Ireland and the UK we convert swap yields to a 360 day year instead of 365, which is the market convention in these three countries. 18

EMU and European Stock Market Integration

14

about the ability of highly indebted governments to control budget de cits, led to substantial increases in long-term interest rate dierentials among European countries. Growing uncertainty about the future of European monetary integration during this period has been re ected in a general increase in forward interest dierentials vis-a-vis Germany and a corresponding decrease in stock market integration. In 1995, integration weights gradually rise again to peak at the end of the sample. Overall, there has been a clear increase in the level of integration over the sample period, suggesting that, as the event of EMU was becoming more certain and expected hedging costs of intra-European currency risk were falling, European investors were gradually increasing their cross-border equity holdings, giving rise to an increase in the relative exposure of local markets to pan-European market risk. Among the countries with a full sample, both the UK and the Netherlands begin with relatively high levels of integration. This is reasonable, given the international character of these economies, with a large number of multinational corporations, international nancial markets and relaxed rules with respect to foreign investment. However, unlike the UK, the integration measure of the Netherlands does approach one at the end of the sample, re ecting the decision by the Netherlands to enter monetary union. Panel B of Table I reports residual analysis of the model. The pricing error for the EU-12 excess return equation is only 0.003% per week and the R2 is 3.19%, whereas for the excess currency return equation the pricing error is -0.002% and the R2 is 6.87%. These R2 s are only slightly lower than those from unrestricted regressions (not reported), suggesting that only a very small amount of predictability has been lost given the restrictions we impose. In columns 4-7 of Panel B we report speci cation tests on the standardized and squared standardized residuals of both the EU market equity return and currency risk model. We cannot reject the null hypothesis that the standardized residuals and the squared standardized residuals are serially uncorrelated. Furthermore, using the positive and negative size and sign bias tests of Engle and Ng (1993) that search for asymmetries in conditional variances, we cannot reject the null hypothesis that the symmetric GARCH model ts the data well. However, normality is rejected for the excess currency return residuals, justifying the use of QML inferential procedures. Next we report Wald tests of the null hypothesis that the price of risk is constant. The null hypothesis is rejected for both the EU-12 excess return model and the excess currency return model. Finally we report orthogonality tests of the residuals against the instruments. For this test we also consider orthogonality of the residuals to a set of global instruments: the world stock market dividend yield in excess of the

EMU and European Stock Market Integration

15

one-month euro-dollar deposit rate, the change in the term spread, de ned as the yield on ten-year benchmark US government bonds and the one-month euro-dollar deposit rate, the change in the default spread, de ned as the yield on US AAA corporate bonds minus the US ten-year bond yield, and the change in the one-month euro-dollar deposit rate. These instruments are the counterparts at the global level of the EU-wide and country speci c instruments used in the estimation. Previous studies on international asset pricing have utilized similar instruments (Harvey (1991), Bekaert and Harvey (1995), De Santis and Gerard (1997)). In an attempt to increase the power of these tests we make the global instruments orthogonal to the EU instruments by regressing the global instruments on the EU instruments and using the residuals from this regression as the new orthogonal instruments. Both sets of residuals are orthogonal to EU and global instruments. With respect to the individual countries, the pricing errors and R2 for our weekly data are very reasonable when compared to the corresponding gures that use monthly data (see, for example De Santis and Gerard (1997) and Harvey (1991)). In seven out of eleven cases we reject the null hypothesis of normally distributed errors. We reject the null hypothesis of no serial correlation of the squared standardized residuals at the 5% level in two cases. The standardized residuals are serially correlated in four cases. In all the local markets we cannot reject the null hypothesis that the GARCH model ts the data well according to the Engle and Ng (1993) test reported in column 7. The next column reports a Wald test of the null hypothesis of a constant price of local risk. The null hypothesis is rejected in every case except Austria. The nal three columns report orthogonality tests. With the exception of Ireland, all residuals are orthogonal to the local instruments at the 5% level of signi cance. With respect to the EU-12 instruments, all residuals are orthogonal at the 5% level except Belgium and Spain. Finally, with respect to the global instruments, the residuals are all orthogonal at the 5% level. Overall, we can reject orthogonality only in 3 out of 37 cases, suggesting that the empirical asset pricing model is well speci ed.

6.1 Sensitivity to Alternative Speci cations of the Model

This section addresses ve potential problems with the estimated model. First, it is possible that, rather than capturing the movement towards EMU, forward interest dierentials simply proxy for some world wide convergence in interest rates based on the synchronization of business cycles and/or monetary policy across the world (interest rate coupling phenomenon). In this case, it is possible that our analysis captures nothing more than common

EMU and European Stock Market Integration

16

worldwide (co-)movements in interest rates and has nothing to do with convergence towards EMU. We do not think this is a realistic possibility because worldwide interest rate couplings are usually followed by decouplings, i.e. they are not permanent. By contrast, the convergence in interest rates in Europe is perceived as permanent. Yet, despite our reservation, to assess whether forward interest rate dierentials contain information about EMU integration in addition to general world information, we compute an orthogonalized forward interest dierential for each local market, which is the residual term from a regression of the local market forward interest dierential vis-a-vis Germany on the German forward interest dierential vis-a-vis the US. The residual term from this regression should contain information related to EMU only. We then reestimate the model with the orthogonal forward interest dierential. Second, the model estimated above imposes the assumption that the price of market risk is positive. De Santis and Gerard (1997) show that in a model of full integration this restriction is rejected by the data. To assess whether the assumption of a positive price of risk matters, we simply omit the exponentiation of the prices of EU-12 market risk and local risk and reestimate the model. Third, given that some of the markets that we consider are small, nonsynchronous trading may be a problem that induces autocorrelation in excess returns. In this case the lagged excess return should not be included as an instrument for the price of local risk, since autocorrelation would not be related to time-variation in risk and required return. Hence, we reestimate the model omitting the lagged excess return and including a rst-order moving average term in the residuals. Fourth, the integration weight may be aected by factors other than the spread. Therefore, we include the local instruments in the spread equation and examine whether they are also important in determining the extent of integration. Fifth, our model leaves out the in uence of world factors on expected returns. If the correct model is integration in a broader world market portfolio, then restricting the model to the EU market ignores the covariance of the local market with the rest of the world (see Chan et al. (1992)). Over the period 1991-1998, the market weights of the rest of the world in the global portfolio decreased by about four percentage points, which could result in the EU risk factors appearing to become more important relative to own variances for our sample of countries. In this case the increase in the degree of integration could re ect the increase in the EU market weight in the world portfolio rather than an increase in the importance of the EU market risk premium. To assess this possibility we weight the EU market risk

EMU and European Stock Market Integration

17

premium with the EU market value as a proportion of total world market capitalization at each point in time and reestimate the model. We provide evidence regarding the performance of each of the ve models above in form of nonnested hypotheses tests. Table 3 reports J-tests of the Davidson and MacKinnon (1981) type. The tests are performed in the following way: Assume under the null hypothesis that the model estimated is H0 : r0;t = f ( 00 Xt;1 ) + "0;t : The alternative model is H1 : r1;t = f ( 10 Zt;1 ) + "1;t , where f (:) denotes a (possibly nonlinear) function, X and Z are the vectors of regressors and 0 and 1 are the corresponding parameter vectors under the null and the alternative hypothesis respectively. We test H0 against H1 by running a regression of the residuals of the model under H0 on the tted values of the model under H1 ; "b0;t = rb1;t + ut ; and testing whether = 0 with a t-test. In testing the null hypothesis that our model speci cation is correct against an alternative speci cation, four possibilities can occur: reject both, neither, or either one of the two hypotheses. Hence, we repeat the J-test by reversing the null and the alternative hypothesis: columns Jn;0 report the results of the reverse test that model n cannot be rejected against our model (model 0) speci cation. The test results indicate that, with the exception of Austria, Finland and Ireland (all with small samples) where model 0 can be rejected against model 1 ans Spain, where model 0 can be rejected against model 4, the test is inconclusive, i.e. neither model is rejected or model 0 cannot be rejected. Overall, out of a total of 55 cases of model comparisons, there is evidence against our model speci cation in only four cases. It should be noted, however, that in three of these cases (Austria, Finland, Ireland) the sample is very small, invalidating the reliability of the test statistic. Excluding these countries, we nd evidence against our model speci cation in only 2.5% of the cases.

6.2 A Model of Full Integration: How Large are the Pricing Errors?

In this sub-section we consider the performance of the model relative to one that assumes the local markets are completely integrated into the EU12. The purpose of this exercise is to illustrate the potential pitfalls of erroneously assuming market integration. In the new model we impose the estimated prices of EU-12 market risk and currency risk and the estimated EU-12 conditional variance on the local markets, whilst omitting the local price of risk. Table IV reports some statistics from this model for each local market. Columns two and three report the average pricing error and the R2 . In

EMU and European Stock Market Integration

18

most cases the pricing errors are larger and the R2 s are smaller than in the partially integrated models. The average pricing error across all markets is 0.237% and the R2 is 1.37% for the fully integrated model, compared to 0.162% and 2.46%, respectively, for the partially integrated model. Clearly, the partially integrated model captures signi cantly more variation in excess returns than a fully integrated model. The extent of these dierences and the resultant impact they would have in any decision making scenario is illustrated in column four, which reports the annualized dierences in expected returns between the model with partial integration and the model with full integration. These dierences range from -0.008% per annum in Austria to over 12% per annum for Portugal and are statistically signi cant in eight of the eleven cases. This indicates that wrongly assuming full integration of stock markets leads to a signi cant underestimation of the cost of capital. For countries which have been struggling to meet EMU entry criteria the dierences in expected returns between the two models tend to be larger, supporting the notion that the cost of capital in less integrated markets is higher. Table IV also reports orthogonality tests of the fully integrated models' errors with respect to the EU-12 instruments, the local instruments and the orthogonalized global instruments. Whilst earlier, in Table I, we reported three rejections of orthogonality in the partially integrated model, using the same signi cance level for the fully integrated model we now reject thirteen times, six of these with respect to local instruments. The nal column of table IV reports a likelihood ratio test of the fully integrated model's restrictions on the partially integrated model, i.e. that g0 = g1 = 0. With the exception of Italy and Netherlands, the restrictions are rejected.

7 Summary and Concluding Remarks The paper examines whether or not the formation of the European Economic and Monetary Union and the adoption of the euro has led to increased integration of European stock markets. We argued that one major consequence of the single currency in Europe was to eliminate certain implicit and explicit restrictions that are related to the currency composition of investors' portfolios. In the context of models of partial segmentation, the reduction in the cost of hedging currency risk and avoiding currency restrictions leads to an increase in cross-border equity holdings, thus, decreasing home equity bias. For a given country, stock market integration is expected to be higher, the higher the probability of the country joining EMU and the closer the calendar date to the time of launch of the single currency, as the expected

EMU and European Stock Market Integration

19

cost of hedging currency risk and avoiding currency restrictions decreases. This in turn leads to an increase in cross-border equity holdings, increasing the relative weight of EU-wide risk in required returns. We discovered that in the 1990s the degree of integration of local markets with the EU market is negatively associated over time with the forward interest rate dierential vis-a-vis Germany. When this interest rate dierential shrinks in 1997 and 1998, the markets converge towards full integration, that is, expected returns are increasingly determined by EU-wide market risk and less by local risk. The model outperforms signi cantly a model that assumes full integration throughout the 1990s in terms of pricing errors and goodness of t measures, suggesting that a partial integration model provides a better speci cation of European risk premia. Finally, the model is robust to a number of alternative speci cations such as the non-negativity constraint on the prices of market and local risk, the empirical proxy for the degree of integration, the in uence of missing global risk factors, and the presence of autocorrelation in the residuals of the return equations.

EMU and European Stock Market Integration

20

References Adler, Michael, and Bernard Dumas, 1983, International portfolio selection and corporation nance: A synthesis, Journal of Finance 46, 925-984. Baba, Yoshihisa, Robert F. Engle, Dennis F. Kraft, and Kenneth F. Kronner, 1989, Multivariate simultaneous generalized ARCH, Working Paper, University of California, San Diego, California. Bekaert, Geert, and Campbell R. Harvey, 1995, Time-varying world market integration, Journal of Finance 50, 403-444.. Black, Fisher, 1974, International capital market equilibrium with investment barriers, Journal of Financial Economics 1, 337-352. Blake, David, Bruce N. Lehman, and Allan Timmermann, 1997, Performance measurement using multiple asset class portfolio data: a study of UK pension funds, CEPR Discussion Paper No. 1618, Centre for Economic Policy Research, London. Bollerslev, Tim, Robert F. Engle, and Jerey M. Wooldridge, 1988, A capital asset pricing model with time-varying covariances, Journal of Political Economy 96, 116-131. Bollerslev, Tim, and Jerey M. Wooldridge, 1992, Quasi-maximum likelihood estimation and inference in dynamic models with timevarying covariances, Econometric Reviews 11, 143-172. Campbell, John Y., 1987, Stock returns and the term structure, Journal of Financial Economics 18, 373-400. Chan, K.C., Andrew Karolyi and Rene M. Stulz, 1992, Global nancial markets and the risk premium on U.S. equity, Journal of Financial Economics 32, 137-168. Cooper, Ian and Evi Kaplanis, 1999, Partially segmented international capital markets and international capital budgeting, Journal of International Money and Finance (forthcomming).

EMU and European Stock Market Integration

21

Davidson, R. and J.G. MacKinnon, 1981, Several tests for model speci cation in the presence of alternative hypotheses, Econometrica 49, 781-793. De Santis, Giorgio and Bruno Gerard, 1997, International asset pricing and portfolio diversi cation with time-varying risk, Journal of Finance 52, 1881-1912. Dumas, Bernard, and Bruno Solnik, 1995, The world price of foreign exchange risk, Journal of Finance 50, 445-479. Engle, Robert F., and Victor K. Ng, 1993, Measuring and testing the impact of news on volatility, Journal of Finance 48, 1749-1778. Errunza, Vihang, and Etienne Losq, 1985, International asset pricing under mild segmentation: Theory and test, Journal of Finance 40, 105-124. Ferson, Wayne E., and Campbell R. Harvey, 1993, The risk and predictability of international equity returns, Review of Financial Studies 6, 527-566. Ferson, Wayne E., and Campbell R. Harvey, 1994, Sources of risk and expected returns in global equity markets, Journal of Banking and Finance 18, 775-803. Hardouvelis, Gikas A., Doncheol Kim, and Thierry Wizman, 1996, Intertemporal asset pricing models with and without consumption: An empirical evaluation, Journal of Empirical Finance 3, 267-301. Harvey, Campbell R., 1991, The world price of covariance risk, Journal of Finance 46, 111-158. Lewis, Karen, 1999, Trying to explain home bias in equities and consumption, Journal of Economic Literature 37, 571-608. Licht, Amir N., 1998, Stock market integration in Europe, CAER II Discussion Paper No. 15, Harvard Institute for International Development, Cambridge, MA. Merton, Robert, C., 1980, On estimating the expected return on the market: An exploratory investigation, Journal of Financial Economics 8, 323-361.

EMU and European Stock Market Integration

22

Solnik, Bruno, 1974, An equilibrium model of the international capital market, Journal of Economic Theory 8, 500-524. Stehle, Richard, 1977, An empirical test of the alternative hypotheses of national and international pricing of risky assets, Journal of Finance 32, 493-502. Stulz, Rene M., 1981, On the eects of barriers to international investment, Journal of Finance 36, 923-934. Stulz, Rene M., 1999, Globalization of equity markets and the cost of capital, NBER Working Paper No. 7021. White, Halbert, 1980, A heteroskedasticity-consistent covariance matrix estimator and direct test for heteroskedasticity, Econometrica 48, 817-838.

Table I Time-Varying Integration and Expected Returns Panel A reports parameter estimates of model (2)-(9). The second and third rows report estimates of the mean return in the EU-12 stock (EU) and currency (C) equations. is the vector of estimated parameters relating returns to EU instruments (XEU t;1 ) as follows: 0 : constant, 1 : dividend yield in excess of one month interest rate, 2 : change in term spread between ten-year and one month interest rates, 3 : default spread, 4 : change in one month interest rate. is the vector of estimated parameters relating returns to local instruments (XLt;1 ) as follows: 0 : constant, 1 : dividend yield in excess of one month interest rate, 2 : change in term spread between ten-year and one month interest rates, 3 : change in one month interest rate,

4 : local market excess return. All instrumental variables enter with one lag. Panel B reports residual analysis and tests of the model. P.E. reports the average weekly percentage pricing error, R2 is the coecient of determination; N is the Bera-Jarque test for normality of the standardized residuals, H is a test statistic for 4th order serial correlation of the squared standardized residuals, S.C. is a test statistic for 4th order serial correlation of the standardized residuals, EN is the Engle-Ng joint test for asymmetries in conditional volatility. W is a robust Wald test of the null hypothesis that the price of local risk is constant ( i;k = 0, k = 1; 2; 3; 4): OEU ;OW ; and OL are Wald tests of orthogonality of the residuals to the EU, World and local instruments, respectively. The tests are distributed 2 (5) except for Ireland where the test for orthogonality with respect to the local instruments is distributed 2 (4). ( ;y ) indicates statistically signi cant at the 1% (5%, 10%) level, respectively. Standard errors are in parentheses and probability values are in square brackets. Country codes are as follows: AU: Austria, BL: Belgium-Luxembourg, FN: Finland, FR: France, GE: Germany, IR: Ireland, IT: Italy, NL: Netherlands, PO: Portugal, SP: Spain, UK: United Kingdom. Data are weekly from 6:7:91 to 26:6:98, except for the following countries: Portugal and Austria (13:1:95 to 26:6:98), Finland (25:10:96 to 26:6:98) and Ireland (9:8:96 to 26:6:98).

23

Panel A: Model Estimates of Mean Return Equation

0

1

2

3

1

2

3

0.177 ;9.435 0.5959 EU 1.109 (0.28) (0.07) (0.39) (1.39) 150.781 130.041 C -13.605 30.821 (8.95) (2.01) (37.41) (16.29)

0

AU BL FN FR GE IR IT NL PO SP UK EU C AU BL FN FR GE IR IT NL PO SP UK

:

(0 10)

4

g1

-9:321y 0.5679 16.230 -6.041 ;98.145 (4.19) (14.60) (18.07) (5:17) (83.46) -12:407 ;4.614 1.979 -0.161 -21.718 (0.48) (2.23) (8.04) (1:05) (2.29) ;6:640 1.662 -0.139 -7.689 -8 :537 (0.59) (6.62) (19.41) (1:38) (2.90) -10:327 0:501 ;0.985 ;0.159 -39.174 (0:25) (18.32) (2:18) (0.71) (0.67) 1:023 1.765 5.409 ; 19.646 -12:979 (0:15) (2.59) (4.31) (1:83) (14.69) 0.129 2.241 91.479 -16:127 (2.69) (10.27) (30.07) (4:80) 1.451 4.065 ; 0:224 -10.178 -8:072 (0.265) (0.76) (100.3) (2:29) (0:75) 18.001 ;5.703 -6.272 -12.056 -13:542 (13.26) (3.21) (12.51) (0:80) (2.64) 0.540 -1.210 -0.012 -3.836 -15:152 (0.08) (1.92) (2.44) (18.10) (2:52) -12:966 0.219 3.665 3 : 673 56 : 917 (0.06) (1.68) (1.85) (7:33) (2:29) -30.773 -18:096 -0.049 -4.462 -4.458 (0.04) (1.02) (0.96) (5.33) (1:45) Panel B: Residual Analysis R2 (%) N 2 (2) H 2 (4) S.C. 2 (4) EN2 (3) W 3.19 4:49 1:47 6:67 3:63 92:73 [0:11] [0:83] [0:15] [0:30] [0:00] 6.87 4761:8 1:01 1:80 1:92 313:78 [0:00] [0:91] [0:77] [0:59] [0:00] 1.01 7.10 23.88 13.34 4.69 1:29 [0.03] [0.00] [0.01] [0.20] [0:86] 1.89 1.83 9.23 9.55 1.16 27 :31 [0.40] [0.06] [0.05] [0.76] [0:00] -0.01 20.87 3.36 4.42 1.74 13 :26 [0.00] [0.50] [0.35] [0.63] [0:01] 2.88 1.03 9.18 1.22 3.59 25 :54 [0.59] [0.06] [0.87] [0.31] [0:00] 1.97 8.96 9.31 3.15 1.95 50 :27 [0.01] [0.05] [0.53] [0.58] [0:00] -1.15 0.33 6.42 1.57 1.49 9:85 [0.84] [0.17] [0.81] [0.68] [0:02] 2:09 4.35 3.37 6.18 3.22 59 :78 [0:11] [0.50] [0.19] [0.36] [0:00] 3:01 76 :77 7.05 9.55 1.20 12 :59 [0:00] [0.13] [0.05] [0.75] [0:01] 1:89 67.19 1.06 10 :65 3.47 43 :41 [0:00] [0:90] [0.03] [0.32] [0:00] 1.44 6.92 2.10 6:61 3.66 82 :35 [0:03] [0.72] [0:16] [0.30] [0:00] 5.98 1.59 8:30 9.31 4.80 76 :10 [0.45] [0:08] [0.05] [0.18] [0:00]

-4.654 (8.85) 4.745 (0.25) 3.678 (0.38) 2.591 (0.71) 4.872 (0.31) 0.715 (0.85) 2.442 (2.559) 3.621 (0.41) 5.398 (0.39) 1(0:974 :34) 2.400 (0.17)

P.E.(%) ;(00::09) 003 ;(00::01) 002 0:110 (0:17) 0(0:122 :08) 0(0:414 :35) 0(0:031 :11) 0(0:025 :09) 0(0:506 :23) 0(0:106 :18) 0(0:113 :09) 0(0:202 :16) 0(0:045 :13) ;0:116

4

-8.358 (1.33) 24.754 (40.84)

24

OEU 4:70 [0:45] 0:58 [0:98] 3:52 [0:62] 11 :59 [0:04] 3:93 [0:55] 0:96 [0:96] 2:57 [0:76] 6:72 [0:24] 6:73 [0:24] 7:25 [0:20] 7:64 [0:18] 11[0::06 05] 8:46 [0:13]

OW 4:15 [0:53] 5:42 [0:37] 5:79 [0:33] 5:21 [0:39] 7:52 [0:18] 1:53 [0:91] 7:39 [0:19] 8:78 [0:12] 3:99 [0:55] 6:49 [0:17] 4:55 [0:47] 6:27 [0:28] 7:99 [0:15]

OL 2:66 : 6[0::15 29] 3[0::94 56] 4[0::12 53] 1[0::56 91] y 8[0::32 08] 6:14 [0:29] 7:47 [0:11] 5:37 [0:37] 2:44 [0:78] 8:47 [0:13]

[0 75]

Table II Robustness of Results: Sensitivity to Alternative Speci cations The table reports a series of non-nested hypotheses tests of the model against ve alternative models. Model 0 is the underlying model for local returns. Model 1 uses a measure of the forward interest dierential that is orthogonal to the US forward interest dierential against Germany. Model 2 does not impose the restriction that the prices of market risk and local risk are positive. In Model 3 we exclude the lagged return from the instruments list and assume an MA(1) process for the residuals instead. In Model 4, the degree of integration is modelled as a function of the spread and the instrument set XLt;1 . In Model 5, the EU market risk premium, hEU;i;t EU;t;1 , is weighted with the world market capitalization of EU-12. We perform J-tests of the Davidson and MacKinnon (1981) type. Each column J0;m reports J-tests of the null hypothesis that model 0 is true against the alternative hypothesis that model m is true. For each model we repeat the J-test by reversing the null and the alternative hypothesis (column Jm;0 ). Heteroskedasticity consistent, marginal signi cance levels are given in square brackets under the test statistics. Signi cance of the test statistic Jn;m means that model n can be rejected against the alternative model m: ( ;y ) indicates statistically signi cant at the 1% (5%, 10%) level, respectively. Model 1 Model 2 Model 3 Model 4 Model 5 Country J0;1 J1;0 J0;2 J2;0 J0;3 J3;0 J0;4 J4;0 J0;5 J5;0 AU 6:71 1:35 ;0:47 2:30 0:85 ;0:87 ;0:16 2:38 0:11 2:61 [0:00] [0:17] [0:02] [0:39] [0:02] [0:91] [0:01] [0:63] [0:38] [0:87] BL 0:75 0:71 1:34 4:06 0:86 1:02 1:53 0:57 ;0:06 0:99 [0:45] [0:47] [0:18] [0:00] [0:39] [0:31] [0:13] [0:56] [0:32] [0:95] FN 2:07 0:59 0:37 1:07 0:87 0:73 1:11 0:04 1:19 1:46 [0:04] [0:55] [0:70] [0:29] [0:38] [0:46] [0:27] [0:96] [0:23] [0:14] 1:75y 2:22 1:56 2:20 FR 1[0::31 1 : 40 1 : 19 3 : 68 0 : 63 2 : 08 19] [0:16] [0:23] [0:00] [0:52] [0:04] [0:08] [0:03] [0:12] [0:03] GE 0[0::62 0 : 33 0 : 85 3 : 16 0 : 34 0 : 78 0 : 50 0 : 50 0 : 28 0:77 53] [0:74] [0:40] [0:00] [0:73] [0:43] [0:62] [0:62] [0:77] [0:44] IR 2[0:33 1:27 2[0:51 2[0:50 2[0:91 2[0:09 0:68 0:51 2:44 2[0:31 :02] [0:20] :01] :01] :00] :04] [0:49] [0:61] [0:01] :02] IT 1:29 1:44 1:46 5:09 0:72 2:44 1:60 1:83y 1:79y 2:03 [0:19] [0:15] [0:14] [0:00] [0:47] [0:01] [0:11] [0:07] [0:07] [0:04] NL 0:78 1:02 0:39 2:92 0:86 1:16 1:34 0:78 1:02 0:99 [0:43] [0:30] [0:69] [0:00] [0:39] [0:24] [0:18] [0:43] [0:30] [0:32] PO 1:12 1:12 1:16 1:47 0:70 1:11 0:82 0:36 0:56 0:64 [0:26] [0:26] [0:24] [0:14] [0:48] [0:26] [0:41] [0:72] [0:57] [0:52] SP 0:68 ;0:29 1:31 2:41 0:57 3:41 2:22 1:07 ;0:38 ;0:14 [0:49] [0:19] [0:02] [0:57] [0:00] [0:03] [0:28] [0:77] [0:70] [0:88] UK 0:54 0:56 0:27 3:38 1:03 2:36 1:43 0:87 0:54 0:71 :

[0 58]

:

[0 57]

:

[0 78]

:

[0 00]

25

:

[0 29]

:

[0 02]

:

[0 15]

:

[0 38]

:

[0 59]

:

[0 47]

Table III A Model of Full Integration This table reports diagnostic tests from estimating a model that assumes the countries in the sample are completely integrated into the EU12. Column 2 reports the average weekly percentage pricing error. Column 3 reports the R2 : Column 4 reports the annualized dierence in percentage points between expected returns of the model with partial integration and expected returns of the model with full integration (E.R. Dif.). Columns 5, 6 and 7 report Wald type orthogonality tests of the residuals with respect to three information sets: the EU-12 instruments (OEU ), the local instruments (OL ) and (orthogonalized) global instruments (OW ). The tests are distributed 2 (5) except for Ireland where the test for orthogonality with respect to the local instruments is distributed 2 (4). Column 8 reports a likelihood ratio test of the fully integrated model against the underlying partial integration model. The restriction tested is g0 = g1 = 0 and the test statistic is distributed 2 (2): Standard errors are in parentheses and probability values are in square brackets. ( ;y ) indicates statistically signi cant at the 1% (5%, 10%) level, respectively. Data are weekly from 6:7:91 to 26:6:98, except for the following countries: Portugal and Austria (13:1:95 to 26:6:98), Finland (25:10:96 to 26:6:98) and Ireland (9:8:96 to 26:6:98). Country P.E. (%) R2 (%) E.R. Dif. (%) AU 0(0:110 0.67 ;[00::92] 008 :13) BL 0(0:193 2.12 1[0:212 :08) :00] y FN 0(0:594 0.11 8:[0561 :35) :00] FR 0:085 1.12 0:816 (0:11) [0:00] GE 0:112 0.55 1:075 (0:09) [0:00] IR 0(0:551 0.08 0 :089 :22) [0:85] IT 0(0:034 1.72 0[0:045 :18) :86] NL 0:(0230 2.22 2 : 894 :09) [0:00] PO 0(0:440 0.52 12[0::343 :16) 00] SP 0:161 3.71 2:987 (0:13) [0:00] UK 0:098 2.27 10:508 :

(0 11)

:

[0 00]

26

OEU

OL

OW

3[0:704 1[0:910 5[0:308 :59] :86] :38] 14:096 13[0:140 9 : 180 :02] [0:01] [0:10] 6[0:521 6 : 013 14 :666 :26] [0:30] [0:01] 1:282 5:995 1:702 [0:93] [0:31] [0:89] 4:479 5:077 8:228 [0:48] [0:41] [0:14] 7[0:185 9 : 291 10 :165y :20] [0:05] [0:07] 3[0:258 3[0:615 7[0:497 :66] :61] :19] 17:794 15:786 12[0::098 03] [0:00] [0:01] 13:308 10:541y 12[0:434 :03] [0:02] [0:06] 4:067 9:870y 7:644 [0:54] [0:08] [0:18] 2:898 21:113 6:329 :

[0 72]

:

[0 00]

:

[0 28]

LR Test 5.40y 9.21 11.10 4.68y 6.20 50.06 2.80 3.12 7.22 9.32 71.60

Figure 1: Stock Markets' Cross-Correlation with EU-12 and Forward Interest Differentials The Figure displays on the left scale a 52-week rolling estimate of the average crosscorrelation coefficient between local stock returns and the EU-12 market return. Individual cross-correlations of the 12 countries with the EU-12 market return are equally weighted. The right scale of the Figure displays the average forward interest differential vis-a-vis Germany for the seven countries for which data on interest rate swaps are available over a longer sample (Belgium - Luxembourg, France, Germany, Italy, Netherlands, Spain, U.K.) as a ratio over the German long bond yield. The forward interest differential contains information about the expected eight-year interest rate differential vis-a-vis Germany in two years' time. The German forward interest differential is calculated vis-a-vis the ECU.

Figure 2: Estimates of Integration Weights Panels A through K plot the estimated integration weights, f , for each of the eleven countries, estimated from a conditional asset pricing model of local excess stock returns. The timevarying degree of integration is estimated as a function of the forward interest differential vis-à-vis Germany and measures the relative importance of EU-wide relative to countryspecific risk. The model accounts for intra-European currency risk, time-varying quantities and prices of risk. The latter are conditioned on EU-wide and country-specific information variables.

Our partners will collect data and use cookies for ad personalization and measurement. Learn how we and our ad partner Google, collect and use data. Agree & close