Did the EMS Reduce the Cost of Capital?

Sentana,, Enrique

doi:10.1111/1468-0297.00069

Did the EMS Reduce the Cost of Capital?

Sentana,, Enrique 2002-10-01 00:00:00 Abstract We propose a dynamic arbitrage pricing theory (APT) multi‐factor model with time‐varying volatility for currency, bond and stock returns for ten European countries over 1977–97. We exploit the cross‐sectional dimension of the model to construct world portfolios, which when added to the original list of assets, allow us to develop simple consistent methods of estimation and testing. Our results reject the implicit asset pricing restrictions, and suggest decreases in idiosyncratic exchange rate risk tend to lower the cost of capital, although the effect is small. We assess the potential gains from increased stock market integration. The European Commission argued (1990, p. 82) that ‘potentially the most important source of gains from European Monetary Union comes from the reduction in overall uncertainty EMU might provide’. They went on to contend that a reduction in exchange rate risk would reduce the risk premium, and that this reduction in the cost of capital would stimulate corporate investment. Nevertheless, as the report acknowledged, it is not a priori obvious that intra‐European exchange rate risk should affect the cost of capital. For example, firms might be able to hedge their exchange rate exposure through a variety of financial instruments. Similarly, those who diversify their investments globally may not be affected by idiosyncratic variations in a country's exchange rate. Therefore, a preliminary question that has to answered is whether country‐specific exchange rate factors are priced in the context of asset pricing models applied at the international level; see Stulz (1995) for a recent survey. Although those risks should not be rewarded in a world with complete market integration, the existence of capital controls or other legal impediments to cross‐border investment (such as limitations on the holdings of foreign securities by pension funds and insurance companies), informational asymmetries, illiquid markets, behavioural biases etc. suggest that idiosyncratic exchange rate risk is likely to be priced. Hence, whether or not EMU would reduce the cost of capital, and if so, by how much, is, ultimately, an empirical question. We attempt to throw some light on this issue within the framework of the dynamic version of the arbitrage pricing theory (APT) developed in King et al. (1994)(KSW hereinafter). Specifically, we use monthly data on currency, bond and stock returns for ten European countries over the period 1977–97 to estimate a multi‐factor model with time‐varying volatility in the underlying factors, in which the idiosyncratic components of returns are (almost) uncorrelated across countries, but their correlation structure is arbitrary within each country. From the methodological point of view, we exploit the cross‐sectional dimension of the model to construct diversified portfolios of European and non‐European securities, which when added to the original list of assets, allow us to develop simple consistent methods of estimation and testing. In particular, we employ a GMM estimation procedure that would be efficient under conditional homoscedasticity, but which remains consistent under the more realistic assumption of dynamic heteroscedasticity. Under the null, our model implies that country‐specific risks should not be priced. However as mentioned before, it is likely that these idiosyncratic risk components do indeed affect the required rate of return on bonds and stocks. If country‐specific exchange rate volatility is associated with higher stock returns, then systems that attempt to reduce nominal exchange rate variability, such as the Exchange Rate Mechanism (ERM) of the European Monetary System (EMS), may well reduce capital costs for firms that raise funds by issuing equity. Since the stock market is not the primary source of finance in some of the countries that we look at, Germany being the prime example, we also look at bond returns.1 In either case, though, we would still be looking at only one component of the cost of capital. Since we include currency returns in our set of assets, we can also say something about the effect of the EMS on the riskless interest rate component. However, under a target zone system such as the ERM, the cost of capital would not necessarily fall; for example, some authors such as Batchelor (1985) have argued that a credible system will inevitably increase interest rate volatility since the authorities are forced to defend the currency. If interest rate volatility is also positively associated with risk premia on bond and equity markets, and if ERM membership raised interest rate volatility, then it is theoretically possible that the EMS might even have increased the cost of capital. Again, this is an issue that we shall investigate. More generally, testing the cross‐equation asset pricing restrictions of our basic model enables us to examine the important question of whether European capital markets are integrated. In this respect, it is worth noting that an important indirect effect of EMU, in conjunction with the development of the single market for financial services, should be the elimination of many of the remaining barriers to cross‐border investments in the EU. To gauge the potential gains from increased market integration, we follow Stulz (1999), and compare stock market risk premia under full integration with the risk premia that would prevail in the context of completely segmented markets. The rest of the paper is organised as follows. Section 1 discusses our basic model and its estimation procedure. Section 2 reports the empirical results, and Section 3 contains our conclusions. This is followed by a Data Appendix. 1. Theory and Estimation 1.1. Asset Pricing Model We base our analysis in a world with a large number of countries, and assume that, for each country, there are three representative assets available: a 1‐period local currency deposit with safe gross return , a long‐term default‐free bond portfolio, whose random gross holding return over period t in local currency is , and a stock portfolio, whose random gross holding return in local currency is . Let be the spot exchange rate for country j at the end of period t in terms of the numeraire currency (US $ in our case), and let be the gross return on the safe asset for the US during period t in US$. The excess returns of the three representative assets for each country in terms of the numeraire currency will be given by (1) where and are the (continuously compounded) excess returns for bonds and stocks in local currency.2 Note that, under covered interest parity, is the variation in exchange rates corrected for the forward premium. Let and be the risk premia on currency, bond and stock returns for country j, and let and be the corresponding unanticipated (as of time t−1) components of returns. Given that simply reflects the unexpected variations in exchange rates against the common numeraire currency, and these are so highly correlated across countries, a single‐factor model should provide a realistic representation of their covariance structure. In this respect, note that full European monetary union would imply that would be the same for all member countries. On the other hand, while both bond and stock returns are additionally affected by interest rate movements, it is generally accepted that stock returns are also exposed to other risks. On this basis, we assume the following conditional factor structure for the innovations in returns: (2) or, in matrix notation, , where fet and fit are common factors representing systematic exchange rate and interest rate risks, fmt is a common residual market risk factor, vjet and vjit are exchange and interest rate risks specific to country j, vjmt represents other risks specific to country j stocks, and the βs and δs are the associated factor loadings that measure the sensitivity of the assets to the different factors, which we assume time‐invariant for any given unconditional normalisation of the factors. To guarantee that the are innovations, we assume that common and specific factors are unpredictable on the basis of past information. Following KSW, we also make the assumption that the common factors are conditionally orthogonal to each other, but allow them to have time‐varying conditional variances λet,λit and λmt.3 As for the idiosyncratic terms, which by definition are conditionally orthogonal to ft, we also assume that they are conditionally orthogonal to one another for a given j, with time‐varying conditional variances ωjet,ωjit and ωjmt. But note that since , the idiosyncratic unconditional covariance matrix remains totally unrestricted within a country. Finally, we assume that the idiosyncratic conditional covariance matrix has the approximate zero‐factor structure introduced by Chamberlain and Rothschild (1983), in which vjet,vjit and vjmt may be correlated across countries, but only mildly so in order to guarantee that full diversification applies.4 Under a mild no arbitrage condition, it is possible to prove that there is a stochastic discount factor, ξt, which prices the available assets by discounting their uncertain pay‐offs across different states of the world; see, for example Cochrane (2001). In terms of excess returns, in particular, we will have that (3) If (and only if) the stochastic discount factor contains no idiosyncratic risk, ξt will be correlated with , and only through ft, and the assets’ risk premia can be written as an exact linear combination of the volatility of the common factors, with weights proportional to the corresponding factor loadings. Specifically, (4) where τk (k=e,i,m) are the prices of risk corresponding to each factor, ie the amount of expected return that agents are willing to give away to reduce its variability by one unit. An alternative way of interpreting the above relationship can be obtained by noticing that risk premia are also linear combinations of the asset factor loadings or betas, with weights that are common to all assets. These weights can be understood as the risk premia of three factor mimicking portfolios, ie three unit‐cost, diversified portfolios of risky assets with unit loadings on the common exchange rate, interest rate and residual stock market risk factors respectively, and zero loadings on the others. In this respect, our model coincides with a conditional version of the exact5 APT pricing relationship.6 Importantly, since we explicitly consider currency returns, our model will also hold for local currency excess returns on bonds and stocks, which can be understood as fully hedged returns; see, for example, Bekaert and Hodrick (1992). Furthermore, it will also hold for stock and bond returns measured in a common (diversified) basket currency. Note that, if our asset pricing model is correct, risk prices depend on the factors, not on the assets, since otherwise there would be arbitrage opportunities. Furthermore, the model also implies that country‐specific risks should not be rewarded because they can be diversified away. As we will see in the next section, these fundamental restrictions are the basis of our tests. 1.2. Alternative Hypotheses To assess whether EMU would reduce the cost of capital and, if so, by how much, our most general alternative is given by (5) where and are defined in (4). This equation enables us to test various hypotheses of interest: We may ask if idiosyncratic European exchange rate variability is not priced in bond and stock markets, ie if ρbje = ρsje =0 ∀ j . If so, then it is difficult to see how the EMS could have affected the equity and bond components of the cost of capital by affecting exchange rate volatility. Similarly, testing if ρcje =0 ∀ j is also very interesting, as it would throw some light on the question of whether the observed convergence of short rates in Europe is explicable by the reduction in exchange rate volatility. In both cases, the values of ρaje ( a = c , b , s ) will allow us to measure the effects of eliminating such idiosyncratic risks. If idiosyncratic interest rate volatility is priced, ie ρbji ≠0 or ρsji ≠0, then the EMS might have affected the cost of capital, if it also affected interest rate volatility. It is possible that other sources of idiosyncratic risk are priced in the stock market, ie ρsjm ≠0, which would suggest that European stock markets are not fully integrated. We may also test for European capital market integration in the sense of asking whether the different prices of risk are common across countries. To do so, we can ask whether freeing up the way in which the conditional variances of the common factors affect risk premia can help us to explain returns better. That is, if θbje = θsje = θbji = θsji = θsjm = 0. 1.3. Estimation Method As by construction, the basic model for excess returns that we seek to estimate can be written in compact form as (6) or, in matrix notation, , where (k = e, i, m) is shorthand for τkλkt + fkt = πkt + fkt. Under the assumption of conditional normality, (6) could be estimated for any N countries simultaneously by maximum likelihood (see KSW for details). With three assets per country, and a non‐diagonal time‐varying conditional idiosyncratic covariance matrix, though, this results in a very time‐consuming procedure even for moderately large N. However, estimation could be considerably simplified if we had data on the basis portfolios; see Sentana (1997a). Unfortunately, we do not usually observe directly. Nevertheless, if we construct three diversified (passive) portfolios consisting of currency deposits, bonds and stocks, with excess returns and respectively, a simple method of estimation and testing can still be developed, similar to the one used by Connor and Korajczyk (1988) for the static APT, and more recently by Forni and Reichlin (1998) and Stock and Watson (1998) for macroeconomic time series applications. Given that the idiosyncratic errors are independent in mean from the δs, the approximate zero factor structure assumption on the vjts implies that (7) Therefore, is a full‐rank, time‐invariant transformation of the basis portfolios, whose factor loadings are a linear combination of the corresponding asset factor loadings. Since the scaling of the common factors is free, we can set without loss of generality. Hence, in matrix notation, we can write , with unit lower‐triangular and obviously . If we now add the three portfolios in (7) to the list of 3N assets under consideration, we can factorise the joint likelihood function into the marginal component of , and the conditional components corresponding to all the individual countries, which are given by (8) where the relationship between both sets of parameters is . Let us start with the marginal model for the portfolios. Since the factors are (conditionally) orthogonal, we can decompose the joint log‐likelihood function of (given the past) into the marginal component of plus the conditional of given plus the conditional of given and . This yields (9) In the conditionally homoscedastic case, (9) is a recursive simultaneous equation system, and the parameter estimates are particularly simple to obtain: (a) πe and λe from the OLS regression of on a constant (b) and λi from the OLS regression of on and a constant (c) and λm from the OLS regression of on and a constant. However, there is a generated regressor problem in (c) which affects inferences involving . Nevertheless, consistent standard errors for all the parameters in (9) can be obtained by regarding the above estimation method as GMM based on the just‐identifying moment conditions implicit in (a), (b) and (c) simultaneously. A significant advantage of the GMM framework is that it is easy to see that the estimators for πs and βs remain consistent when fet,fit and fmt follow univariate GARCH processes (and even when the prices of risk are time‐varying), provided that the factor representing portfolios remain contemporaneously uncorrelated, although we must take into account that the residuals will be serially correlated in order to compute consistent standard errors.7 Let us now turn to the conditional models for given . In the conditionally homoscedastic case, ML estimates of can be obtained by using seemingly unrelated regression (SUR) techniques. If the idiosyncratic covariance matrix were block‐diagonal, SUR applied to the 3N assets simultaneously would be equivalent to SUR applied to the three representative assets of each country at a time with as regressors and, in addition, there would be no efficiency loss in estimating the model only for the N countries of interest. Besides, the triangular nature of implies that (iterated) SUR estimates could be obtained simply as follows; see Sentana (1997b): (d) and ωje from the OLS regression of on (e) and ωji from the OLS regression of on and with the residual from (d) as an extra regressor (f) and ωjm from the OLS regression of on and with the residuals from (d) and (e) as extra regressors. Moreover, given the Frisch‐‐Waugh theorem, it is straightforward to prove that the estimates of could be obtained directly if we ran the OLS regressions in (d), (e) and (f) with the orthogonalised portfolios, instead. However, as before, the standard errors associated with and would suffer from a generated regressor bias. Nevertheless, consistent standard errors for all the parameters could be obtained by recasting the above estimation method as GMM based on the just‐identifying moment conditions implicit in (a), (b) and (c) together with (d), (e) and (f) for all countries simultaneously. A significant advantage of the GMM framework is that it is straightforward to conduct inferences involving parameters from different equations that remain valid when some of the assumptions made for estimation, such as block diagonality of the idiosyncratic covariance matrix, do not hold. Furthermore, it is also easy to see that the GMM parameter estimators remain consistent even if ωjet,ωjit and ωjmt follow GARCH processes.8 The different alternative hypotheses discussed above can also be easily tested in this GMM‐regression framework. For instance, to see if idiosyncratic exchange rate variability affects currency risk premia, we can test whether adding the estimated conditional variance of vjet, , as an extra regressor in (6a) improves the explanatory power of the equation. The rationale comes from the fact that (6a) can be rewritten as E. Such a testing procedure also yields as a by‐product an estimate of the coefficient for ωjet under the alternative, ρcje, which is of interest to measure the effects of reducing the variability of idiosyncratic risks. In view of the relatively short number of observations, we choose univariate GARCH (1,1) specifications for the conditional variances of both common and idiosyncratic factors. Such a parametrisation has been found to be a good representation of many financial time series. More efficient conditional variance estimates could be obtained by using a larger information set, or taking into account peso‐type effects. However, note that, even if the GARCH(1,1) model were incorrect, our tests would still be consistent, albeit less powerful, since they are simply rational expectations‐type orthogonality tests, and the GARCH model generates a conditional variance estimate which, conditional on the parameters, is a function of past information. In this respect, it is important to emphasise that our tests will have the correct asymptotic size under the null despite the fact that the estimated conditional variances are generated regressors; see Pagan and Ullah (1988). 2. Empirical Application 2.1. Data We estimate the model described above using monthly data for currency, bond and stock returns (in percentage terms) on ten European countries between October 1977 and October 1997 (ie 241 observations), with the first observation used for initialising the conditional variance recursions. Details of data sources and transformations can be found in the Appendix. The ten countries are Belgium, Denmark, France, Germany, Italy, the Netherlands, Spain, Sweden, Switzerland and the UK. Importantly, note that the returns on both bonds and particularly stocks for each country correspond already to a well‐diversified basket of domestic assets. Since it is crucial that the aggregate portfolios contain the larger non‐European countries as well, we also have data on Australia, Canada, Japan and the US. In this respect, we consider equally weighted world currency, bond and stock returns as our set of portfolios in (7), which are linearly related to the OLS representing portfolios obtained from (6) under the assumption that βajk = 1(a = c, b, s; k = e, i, m) for all the countries for which we have data; see Sentana (2002).9 It is important to mention that Sweden and Switzerland have never belonged to the EMS, although their currencies were unofficially pegged to the DM for part of the sample. The UK only entered the ERM in October 1990, to leave in September 1992 together with Italy, who rejoined the system in November 1996. Finally, Spain, who joined in June 1989, was forced to realign the central parity of its currency four times between September 1992 and March 1995.10 These three countries negotiated a wide margin of ±6% for their bilateral exchange rates, while the other countries, including Italy between 1989 and 1992, participated with bilateral limits of ±2.25% until August 1993, when the bands were widened to ±15%. 2.2. Estimates of the Asset Pricing Model We initially estimated the 195 parameters characterising the model under the null by maximum likelihood as explained in Section 1.3. As a by‐product, we obtained estimates of the conditional variances of common and specific factors. Unfortunately, the computation of consistent standard errors à la Bollerslev and Wooldridge (1992), which avoid the generated regressor problem and are robust to non‐normality, is not an easy task in our case, as it involves the outer product of the gradient and the Hessian matrix of the joint log‐likelihood function for the 33 asset returns. In addition, the outer‐product matrix is singular under some of the alternatives because the number of parameters exceeds the number of observations. For that reason, we only present the GMM estimates discussed in Section 1.3, which turn out to be rather close to the ML ones.11 As for standard errors, we use the Newey and West (1987) formulae with a baseline bandwidth of 6 lags (≃2401/3). Qualitatively similar results are obtained by doubling or halving the lag length. Nevertheless, it is important to emphasise that the finite sample properties of the covariance matrix estimators and associated test procedures in our set‐up are unknown. The results for the subsystem (7) are presented in Table 1. As one would expect from the fact that all returns are US$ denominated, the estimated factor loadings confirm that all three portfolios are positively correlated. Besides, they also confirm that, controlling for exchange rate variations, world bond and stock returns are positively correlated. In this respect, note that since is very close to 1, the interest basis portfolio is hardly distinguishable from an equally weighted average of excess bond returns in local currency. The estimates of the πs, though, are imprecise. The average expected returns on common risks are all positive, but insignificant at conventional levels, with the possible exception of the common interest rate risk if we consider the more relevant one‐sided test.12 Table 1 Prices of Risk and Factor Loadings for Equally‐weighted World Portfolios* World . Common exchange . Common interest . Common stock . Unconditional . portfolio . rate risk . rate risk . market risk . variances . Currencies λe =7.18554 (0.746466) Bonds βbpi =1 λi =1.54193 (0.038047) (0.203489) Stocks βspi =1.33681 βspm =1 λm =11.3880 (0.115252) (0.212905) (3.39880) Unconditional πe =0.00849632 πi =0.219093 πm =0.356913 risk premia (0.198522) (0.126910) (0.243211) World . Common exchange . Common interest . Common stock . Unconditional . portfolio . rate risk . rate risk . market risk . variances . Currencies λe =7.18554 (0.746466) Bonds βbpi =1 λi =1.54193 (0.038047) (0.203489) Stocks βspi =1.33681 βspm =1 λm =11.3880 (0.115252) (0.212905) (3.39880) Unconditional πe =0.00849632 πi =0.219093 πm =0.356913 risk premia (0.198522) (0.126910) (0.243211) Note *Sample period 1977:11‐‐1997:10. GMM estimates of (7): , , , πk= E(πkt), λk= E(λkt)=V(fkt)(k=e,i,m). Newey‐‐West (1987) heteroscedasticity and autocorrelation robust standard errors in parenthesis. Open in new tab Table 1 Prices of Risk and Factor Loadings for Equally‐weighted World Portfolios* World . Common exchange . Common interest . Common stock . Unconditional . portfolio . rate risk . rate risk . market risk . variances . Currencies λe =7.18554 (0.746466) Bonds βbpi =1 λi =1.54193 (0.038047) (0.203489) Stocks βspi =1.33681 βspm =1 λm =11.3880 (0.115252) (0.212905) (3.39880) Unconditional πe =0.00849632 πi =0.219093 πm =0.356913 risk premia (0.198522) (0.126910) (0.243211) World . Common exchange . Common interest . Common stock . Unconditional . portfolio . rate risk . rate risk . market risk . variances . Currencies λe =7.18554 (0.746466) Bonds βbpi =1 λi =1.54193 (0.038047) (0.203489) Stocks βspi =1.33681 βspm =1 λm =11.3880 (0.115252) (0.212905) (3.39880) Unconditional πe =0.00849632 πi =0.219093 πm =0.356913 risk premia (0.198522) (0.126910) (0.243211) Note *Sample period 1977:11‐‐1997:10. GMM estimates of (7): , , , πk= E(πkt), λk= E(λkt)=V(fkt)(k=e,i,m). Newey‐‐West (1987) heteroscedasticity and autocorrelation robust standard errors in parenthesis. Open in new tab The GMM estimates of the factor loadings for each of the three asset classes can be found in Tables 2–4. For clarity of exposition, but without loss of generality, the results for bonds and stocks correspond hereinafter to local currency returns. Since and , the only coefficients affected are those related to the effects of common and specific exchange rate risks. In particular, , , and . Table 2 Factor Loadings for Currency Returns ($)* . Common exchange . Idiosyncratic . Country . rate risk () . variance (ωje) . Belgium 1.26526 0.856507 (0.025079) (0.160474) Denmark 1.22027 0.894656 (0.023217) (0.125631) France 1.20585 0.987137 (0.024049) (0.152557) Germany 1.24651 0.848515 (0.023489) (0.092211) Italy 1.04731 2.17514 (0.047859) (0.606642) Netherlands 1.26122 0.817188 (0.025452) (0.105427) Spain 1.05542 2.40062 (0.049952) (0.602722) Sweden 1.00790 2.87831 (0.056559) (1.06253) Switzerland 1.33764 2.61699 (0.036824) (0.384713) UK 1.02758 4.21787 (0.063866) (0.609914) . Common exchange . Idiosyncratic . Country . rate risk () . variance (ωje) . Belgium 1.26526 0.856507 (0.025079) (0.160474) Denmark 1.22027 0.894656 (0.023217) (0.125631) France 1.20585 0.987137 (0.024049) (0.152557) Germany 1.24651 0.848515 (0.023489) (0.092211) Italy 1.04731 2.17514 (0.047859) (0.606642) Netherlands 1.26122 0.817188 (0.025452) (0.105427) Spain 1.05542 2.40062 (0.049952) (0.602722) Sweden 1.00790 2.87831 (0.056559) (1.06253) Switzerland 1.33764 2.61699 (0.036824) (0.384713) UK 1.02758 4.21787 (0.063866) (0.609914) Note *Sample period 1977:11–1997:10. GMM estimates of (6a): , ωje= E(ωjet)= V(vjet). Newey‐‐West (1987) heteroscedasticity and autocorrelation robust standard errors in parenthesis. Open in new tab Table 2 Factor Loadings for Currency Returns ($)* . Common exchange . Idiosyncratic . Country . rate risk () . variance (ωje) . Belgium 1.26526 0.856507 (0.025079) (0.160474) Denmark 1.22027 0.894656 (0.023217) (0.125631) France 1.20585 0.987137 (0.024049) (0.152557) Germany 1.24651 0.848515 (0.023489) (0.092211) Italy 1.04731 2.17514 (0.047859) (0.606642) Netherlands 1.26122 0.817188 (0.025452) (0.105427) Spain 1.05542 2.40062 (0.049952) (0.602722) Sweden 1.00790 2.87831 (0.056559) (1.06253) Switzerland 1.33764 2.61699 (0.036824) (0.384713) UK 1.02758 4.21787 (0.063866) (0.609914) . Common exchange . Idiosyncratic . Country . rate risk () . variance (ωje) . Belgium 1.26526 0.856507 (0.025079) (0.160474) Denmark 1.22027 0.894656 (0.023217) (0.125631) France 1.20585 0.987137 (0.024049) (0.152557) Germany 1.24651 0.848515 (0.023489) (0.092211) Italy 1.04731 2.17514 (0.047859) (0.606642) Netherlands 1.26122 0.817188 (0.025452) (0.105427) Spain 1.05542 2.40062 (0.049952) (0.602722) Sweden 1.00790 2.87831 (0.056559) (1.06253) Switzerland 1.33764 2.61699 (0.036824) (0.384713) UK 1.02758 4.21787 (0.063866) (0.609914) Note *Sample period 1977:11–1997:10. GMM estimates of (6a): , ωje= E(ωjet)= V(vjet). Newey‐‐West (1987) heteroscedasticity and autocorrelation robust standard errors in parenthesis. Open in new tab Table 3 Factor Loadings for Bond Returns (local currency)* . Common . Common . Specific . Idosyncratic . . exchange rate . interest rate . exchange rate . variance . Country . risk () . risk (βbji) . risk () . (ωji) . Belgium 0.012869 0.746426 −0.010701 1.03904 (0.040965) (0.063506) (0.099932) (0.134862) Denmark −0.031760 1.15055 0.05528 4.34104 (0.052475) (0.098457) (0.135574) (1.02396) France −0.004692 1.02727 0.082677 1.40630 (0.047316) (0.072686) (0.091778) (0.211966) Germany 0.143256 0.929245 0.270939 0.882759 (0.056360) (0.065022) (0.074528) (0.147316) Italy −0.047515 0.962191 0.347770 2.88643 (0.053223) (0.104847) (0.078079) (0.471932) Netherlands 0.069076 0.908067 0.225466 0.863978 (0.042638) (0.059902) (0.077839) (0.153695) Spain −0.086345 0.819648 −0.014769 2.71115 (0.053599) (0.133285) (0.079147) (0.248375) Sweden −0.045976 0.788694 0.060973 2.70035 (0.052172) (0.115046) (0.122362) (0.353565) Switzerland 0.049296 0.590225 0.067478 1.04408 (0.039099) (0.075243) (0.041963) (0.123455) UK 0.056706 1.36035 0.277993 3.94258 (0.056990) (0.109731) (0.092055) (0.564751) . Common . Common . Specific . Idosyncratic . . exchange rate . interest rate . exchange rate . variance . Country . risk () . risk (βbji) . risk () . (ωji) . Belgium 0.012869 0.746426 −0.010701 1.03904 (0.040965) (0.063506) (0.099932) (0.134862) Denmark −0.031760 1.15055 0.05528 4.34104 (0.052475) (0.098457) (0.135574) (1.02396) France −0.004692 1.02727 0.082677 1.40630 (0.047316) (0.072686) (0.091778) (0.211966) Germany 0.143256 0.929245 0.270939 0.882759 (0.056360) (0.065022) (0.074528) (0.147316) Italy −0.047515 0.962191 0.347770 2.88643 (0.053223) (0.104847) (0.078079) (0.471932) Netherlands 0.069076 0.908067 0.225466 0.863978 (0.042638) (0.059902) (0.077839) (0.153695) Spain −0.086345 0.819648 −0.014769 2.71115 (0.053599) (0.133285) (0.079147) (0.248375) Sweden −0.045976 0.788694 0.060973 2.70035 (0.052172) (0.115046) (0.122362) (0.353565) Switzerland 0.049296 0.590225 0.067478 1.04408 (0.039099) (0.075243) (0.041963) (0.123455) UK 0.056706 1.36035 0.277993 3.94258 (0.056990) (0.109731) (0.092055) (0.564751) Note *Sample period 1977:11‐‐1997:10. GMM estimates of (6b): , ωji= E(ωjit)= V(vjit). Newey‐‐West (1987) heteroscedasticity and autocorrelation robust standard errors in parenthesis. Open in new tab Table 3 Factor Loadings for Bond Returns (local currency)* . Common . Common . Specific . Idosyncratic . . exchange rate . interest rate . exchange rate . variance . Country . risk () . risk (βbji) . risk () . (ωji) . Belgium 0.012869 0.746426 −0.010701 1.03904 (0.040965) (0.063506) (0.099932) (0.134862) Denmark −0.031760 1.15055 0.05528 4.34104 (0.052475) (0.098457) (0.135574) (1.02396) France −0.004692 1.02727 0.082677 1.40630 (0.047316) (0.072686) (0.091778) (0.211966) Germany 0.143256 0.929245 0.270939 0.882759 (0.056360) (0.065022) (0.074528) (0.147316) Italy −0.047515 0.962191 0.347770 2.88643 (0.053223) (0.104847) (0.078079) (0.471932) Netherlands 0.069076 0.908067 0.225466 0.863978 (0.042638) (0.059902) (0.077839) (0.153695) Spain −0.086345 0.819648 −0.014769 2.71115 (0.053599) (0.133285) (0.079147) (0.248375) Sweden −0.045976 0.788694 0.060973 2.70035 (0.052172) (0.115046) (0.122362) (0.353565) Switzerland 0.049296 0.590225 0.067478 1.04408 (0.039099) (0.075243) (0.041963) (0.123455) UK 0.056706 1.36035 0.277993 3.94258 (0.056990) (0.109731) (0.092055) (0.564751) . Common . Common . Specific . Idosyncratic . . exchange rate . interest rate . exchange rate . variance . Country . risk () . risk (βbji) . risk () . (ωji) . Belgium 0.012869 0.746426 −0.010701 1.03904 (0.040965) (0.063506) (0.099932) (0.134862) Denmark −0.031760 1.15055 0.05528 4.34104 (0.052475) (0.098457) (0.135574) (1.02396) France −0.004692 1.02727 0.082677 1.40630 (0.047316) (0.072686) (0.091778) (0.211966) Germany 0.143256 0.929245 0.270939 0.882759 (0.056360) (0.065022) (0.074528) (0.147316) Italy −0.047515 0.962191 0.347770 2.88643 (0.053223) (0.104847) (0.078079) (0.471932) Netherlands 0.069076 0.908067 0.225466 0.863978 (0.042638) (0.059902) (0.077839) (0.153695) Spain −0.086345 0.819648 −0.014769 2.71115 (0.053599) (0.133285) (0.079147) (0.248375) Sweden −0.045976 0.788694 0.060973 2.70035 (0.052172) (0.115046) (0.122362) (0.353565) Switzerland 0.049296 0.590225 0.067478 1.04408 (0.039099) (0.075243) (0.041963) (0.123455) UK 0.056706 1.36035 0.277993 3.94258 (0.056990) (0.109731) (0.092055) (0.564751) Note *Sample period 1977:11‐‐1997:10. GMM estimates of (6b): , ωji= E(ωjit)= V(vjit). Newey‐‐West (1987) heteroscedasticity and autocorrelation robust standard errors in parenthesis. Open in new tab Table 4 Factor Loadings for Stock Returns (local currency)* . Common . Common . Common . Specific . Specific . Idiosyncratic . . exchange rate . interest rate . market . exchange rate . interest rate . variance . Country . risk () . risk (βsji ) . risk (βsjm) . risk () . risk (δsji) . (ωjm) . Belgium −0.322000 1.54156 0.869533 0.128528 0.752030 11.0754 (0.1168345) (0.225872) (0.072319) (0.244034) (0.203083) (2.26265) Denmark −0.374027 1.30474 0.642234 −0.247583 0.627680 13.9370 (0.120046) (0.212141) (0.086603) (0.298787) (0.132589) (1.65446) France −0.117952 1.66850 1.19002 0.790233 1.19843 14.4385 (0.183294) (0.300995) (0.069961) (0.378364) (0.268907) (2.19152) Germany −0.239521 1.50536 1.07267 0.774702 0.695732 12.4569 (0.148164) (0.298862) (0.078756) (0.254677) (0.217266) (1.87028) Italy −0.453854 1.37731 1.15685 0.458841 0.575473 30.5493 (0.195227) (0.418395) (0.191447) (0.238861) (0.225265) (4.73570) Netherlands −0.509754 1.56830 0.979499 −0.239149 0.333657 6.80658 (0.143979) (0.227616) (0.064407) (0.176940) (0.165672) (0.932920) Spain −0.264740 1.31521 1.08326 −0.057328 0.724307 19.1805 (0.147007) (0.400857) (0.101797) (0.195961) (0.179790) (2.55937) Sweden −0.493719 1.29952 1.19243 −0.644218 0.570327 21.0370 (0.171656) (0.367956) (0.121713) (0.287745) (0.231128) (2.55007) Switzerland −0.286744 1.56240 0.895178 −0.035626 0.755266 7.66734 (0.146489) (0.260817) (0.079937) (0.137907) (0.158261) (0.856350) UK −0.221163 1.46592 1.01946 −0.122607 0.689536 8.18090 (0.142131) (0.262333) (0.080639) (0.120984) (0.114595) (1.04401) . Common . Common . Common . Specific . Specific . Idiosyncratic . . exchange rate . interest rate . market . exchange rate . interest rate . variance . Country . risk () . risk (βsji ) . risk (βsjm) . risk () . risk (δsji) . (ωjm) . Belgium −0.322000 1.54156 0.869533 0.128528 0.752030 11.0754 (0.1168345) (0.225872) (0.072319) (0.244034) (0.203083) (2.26265) Denmark −0.374027 1.30474 0.642234 −0.247583 0.627680 13.9370 (0.120046) (0.212141) (0.086603) (0.298787) (0.132589) (1.65446) France −0.117952 1.66850 1.19002 0.790233 1.19843 14.4385 (0.183294) (0.300995) (0.069961) (0.378364) (0.268907) (2.19152) Germany −0.239521 1.50536 1.07267 0.774702 0.695732 12.4569 (0.148164) (0.298862) (0.078756) (0.254677) (0.217266) (1.87028) Italy −0.453854 1.37731 1.15685 0.458841 0.575473 30.5493 (0.195227) (0.418395) (0.191447) (0.238861) (0.225265) (4.73570) Netherlands −0.509754 1.56830 0.979499 −0.239149 0.333657 6.80658 (0.143979) (0.227616) (0.064407) (0.176940) (0.165672) (0.932920) Spain −0.264740 1.31521 1.08326 −0.057328 0.724307 19.1805 (0.147007) (0.400857) (0.101797) (0.195961) (0.179790) (2.55937) Sweden −0.493719 1.29952 1.19243 −0.644218 0.570327 21.0370 (0.171656) (0.367956) (0.121713) (0.287745) (0.231128) (2.55007) Switzerland −0.286744 1.56240 0.895178 −0.035626 0.755266 7.66734 (0.146489) (0.260817) (0.079937) (0.137907) (0.158261) (0.856350) UK −0.221163 1.46592 1.01946 −0.122607 0.689536 8.18090 (0.142131) (0.262333) (0.080639) (0.120984) (0.114595) (1.04401) Note *Sample period 1977:11–1997:10. GMM estimates of (6c): , ωjm= E(ωjmt)= V(vjmt). Newey‐‐West (1987) heteroscedasticity and autocorrelation robust standard errors in parenthesis. Open in new tab Table 4 Factor Loadings for Stock Returns (local currency)* . Common . Common . Common . Specific . Specific . Idiosyncratic . . exchange rate . interest rate . market . exchange rate . interest rate . variance . Country . risk () . risk (βsji ) . risk (βsjm) . risk () . risk (δsji) . (ωjm) . Belgium −0.322000 1.54156 0.869533 0.128528 0.752030 11.0754 (0.1168345) (0.225872) (0.072319) (0.244034) (0.203083) (2.26265) Denmark −0.374027 1.30474 0.642234 −0.247583 0.627680 13.9370 (0.120046) (0.212141) (0.086603) (0.298787) (0.132589) (1.65446) France −0.117952 1.66850 1.19002 0.790233 1.19843 14.4385 (0.183294) (0.300995) (0.069961) (0.378364) (0.268907) (2.19152) Germany −0.239521 1.50536 1.07267 0.774702 0.695732 12.4569 (0.148164) (0.298862) (0.078756) (0.254677) (0.217266) (1.87028) Italy −0.453854 1.37731 1.15685 0.458841 0.575473 30.5493 (0.195227) (0.418395) (0.191447) (0.238861) (0.225265) (4.73570) Netherlands −0.509754 1.56830 0.979499 −0.239149 0.333657 6.80658 (0.143979) (0.227616) (0.064407) (0.176940) (0.165672) (0.932920) Spain −0.264740 1.31521 1.08326 −0.057328 0.724307 19.1805 (0.147007) (0.400857) (0.101797) (0.195961) (0.179790) (2.55937) Sweden −0.493719 1.29952 1.19243 −0.644218 0.570327 21.0370 (0.171656) (0.367956) (0.121713) (0.287745) (0.231128) (2.55007) Switzerland −0.286744 1.56240 0.895178 −0.035626 0.755266 7.66734 (0.146489) (0.260817) (0.079937) (0.137907) (0.158261) (0.856350) UK −0.221163 1.46592 1.01946 −0.122607 0.689536 8.18090 (0.142131) (0.262333) (0.080639) (0.120984) (0.114595) (1.04401) . Common . Common . Common . Specific . Specific . Idiosyncratic . . exchange rate . interest rate . market . exchange rate . interest rate . variance . Country . risk () . risk (βsji ) . risk (βsjm) . risk () . risk (δsji) . (ωjm) . Belgium −0.322000 1.54156 0.869533 0.128528 0.752030 11.0754 (0.1168345) (0.225872) (0.072319) (0.244034) (0.203083) (2.26265) Denmark −0.374027 1.30474 0.642234 −0.247583 0.627680 13.9370 (0.120046) (0.212141) (0.086603) (0.298787) (0.132589) (1.65446) France −0.117952 1.66850 1.19002 0.790233 1.19843 14.4385 (0.183294) (0.300995) (0.069961) (0.378364) (0.268907) (2.19152) Germany −0.239521 1.50536 1.07267 0.774702 0.695732 12.4569 (0.148164) (0.298862) (0.078756) (0.254677) (0.217266) (1.87028) Italy −0.453854 1.37731 1.15685 0.458841 0.575473 30.5493 (0.195227) (0.418395) (0.191447) (0.238861) (0.225265) (4.73570) Netherlands −0.509754 1.56830 0.979499 −0.239149 0.333657 6.80658 (0.143979) (0.227616) (0.064407) (0.176940) (0.165672) (0.932920) Spain −0.264740 1.31521 1.08326 −0.057328 0.724307 19.1805 (0.147007) (0.400857) (0.101797) (0.195961) (0.179790) (2.55937) Sweden −0.493719 1.29952 1.19243 −0.644218 0.570327 21.0370 (0.171656) (0.367956) (0.121713) (0.287745) (0.231128) (2.55007) Switzerland −0.286744 1.56240 0.895178 −0.035626 0.755266 7.66734 (0.146489) (0.260817) (0.079937) (0.137907) (0.158261) (0.856350) UK −0.221163 1.46592 1.01946 −0.122607 0.689536 8.18090 (0.142131) (0.262333) (0.080639) (0.120984) (0.114595) (1.04401) Note *Sample period 1977:11–1997:10. GMM estimates of (6c): , ωjm= E(ωjmt)= V(vjmt). Newey‐‐West (1987) heteroscedasticity and autocorrelation robust standard errors in parenthesis. Open in new tab The coefficient on the common exchange rate factor is positive and statistically significant for every currency return (Table 2). Turning to the effect of the common exchange rate on bond returns, note that the coefficients are both positive and negative, but only significantly positive for Germany. This could be because for this country, whose central bank was traditionally concerned about imported inflation, a dollar depreciation leads to lower inflation and, hence, higher bond returns. Table 3, column 2, shows that, in all cases, increases in the common interest rate factor (ie the ‘world’ bond return) are associated with higher European bond returns. Not surprisingly, this effect is always statistically significant. Finally, the idiosyncratic exchange rate factors are significantly positive for Germany, Italy, the Netherlands and the UK. Again, the intuition would be that a local currency appreciation leads to lower inflation and higher bond returns. We next examine the factor loadings for stock returns (Table 4). The coefficients on the common exchange rate are negative in all cases, and significantly so for many countries. Hence, local currency stock returns in Europe suffer when the dollar depreciates. On the other hand, the coefficients on the common interest rate factor suggests that periods of higher world bond returns would tend to be associated with higher European stock market returns. In addition, the coefficients on the common residual market risk factor are all significantly positive. Turning to the effects of the idiosyncratic factors on stock returns, we can see that the coefficients of country‐specific exchange rate factors are positive in some cases and negative in others. In particular, an idiosyncratic local currency appreciation significantly increases stock returns in France and Germany, while decreasing them in Sweden. The fact that, for a given country, the sensitivity of returns to common and idiosyncratic exchange rate movements is different is likely to reflect the structure of its foreign trade. Finally, the results also suggest that controlling for exchange rate variations, the idiosyncratic terms for bonds and stocks are positively and significantly correlated within countries.13 2.3. Direct Effects of ERM on the Cost of Capital Figure 1a displays the average of the conditional standard deviations of the idiosyncratic exchange rate components in ERM and non‐ERM countries. For these purposes, and in line with the consensus view among both academics and practitioners, we only include Belgium, Denmark, France, Germany and the Netherlands in the ERM bloc. Notice that the temporal evolution in both groups is somewhat similar, with a fairly tranquil period between 1983 and 1992, preceded and followed by more volatile ones. Not surprisingly, the effects of the successive EMS crises that began in September 1992 are noticeable in the two series, since the speculative attacks affected all European currencies apart from the Dutch guilder. Nevertheless, the average level and magnitude of the movements are rather different, being substantially smaller for those countries that did not modify the central parity of their currencies against the ECU. Therefore, there is clear evidence that a credible target zone system does reduce exchange rate volatility. Fig. 1. Open in new tabDownload slide Average Conditional Standard Deviation of (a) Idiosyncratic Exchange Rate Factors and (b) Idiosyncratic Interest Rate Factors Fig. 1. Open in new tabDownload slide Average Conditional Standard Deviation of (a) Idiosyncratic Exchange Rate Factors and (b) Idiosyncratic Interest Rate Factors Figure 1b is the analogous picture for the idiosyncratic interest rate components. Except during the first part of the sample, and a brief period between March and May 1990, average interest rate volatility was actually smaller in ERM countries than in non‐ERM ones. At the same time, it also seems that, during turbulent periods in the foreign exchange market, idiosyncratic interest rate volatility increased more for those countries which insisted in maintaining their currencies within the bands. Hence, our results confirm that the reduction in idiosyncratic exchange rate volatility may sometimes be achieved at the expense of increases in local interest rate volatility. Fig. 2. Open in new tabDownload slide Net Effect (a) of Idiosyncratic Exchange Rate Volatility on Currency Returns; and Interest Rate Volatility on Bond Returns; (b) of Idiosyncratic Exchange Rate; and (c) of Idiosyncratic Exchange Rate and Interest Rate Volatility on Stock Returns Fig. 2. Open in new tabDownload slide Net Effect (a) of Idiosyncratic Exchange Rate Volatility on Currency Returns; and Interest Rate Volatility on Bond Returns; (b) of Idiosyncratic Exchange Rate; and (c) of Idiosyncratic Exchange Rate and Interest Rate Volatility on Stock Returns According to our basic asset pricing model, though, such volatility movements should have no impact on expected returns in a world of complete financial market integration. For that reason, we begin by testing whether, contrary to the theory, idiosyncratic exchange rate and interest rate risks are priced. The results for the pricing of idiosyncratic exchange rate volatility in the currency, bond and stock markets of ERM and non‐ERM countries are presented in Table 5 (rows 1 to 3 of panels A and B, respectively). On the currency markets, the joint Wald tests clearly indicate a rejection of the null hypothesis, both in core ERM countries, and especially non‐ERM ones. As for local currency bond and stock returns, the tests would reject in the case of non‐ERM countries, but not in the case of core ERM ones. Table 5 Joint Tests for Pricing of Idiosyncratic Exchange Rate and Interest Rates Risks* Null hypothesis . Risk . Asset . Joint Wald test . p‐value . A: Core ERM countries exchange rate currencies ($) 12.7396 0.02594 exchange rate bonds (l.c.) 7.8618 0.16402 exchange rate stocks (l.c.) 2.2047 0.82016 ρbji =0 ∀ j interest rate bonds 11.7580 0.03826 ρsji =0 ∀ j interest rate stocks 8.3415 0.13839 B: Non‐ERM countries ∀ j exchange rate currencies ($) 47.4100 0.0 ∀ j exchange rate bonds (l.c.) 11.3007 0.04573 ∀ j exchange rate stocks (l.c.) 16.3413 0.00593 ρbji =0 ∀ j interest rate bonds 17.2587 0.00403 ρsji =0 ∀ j interest rate stocks 11.8358 0.03711 Null hypothesis . Risk . Asset . Joint Wald test . p‐value . A: Core ERM countries exchange rate currencies ($) 12.7396 0.02594 exchange rate bonds (l.c.) 7.8618 0.16402 exchange rate stocks (l.c.) 2.2047 0.82016 ρbji =0 ∀ j interest rate bonds 11.7580 0.03826 ρsji =0 ∀ j interest rate stocks 8.3415 0.13839 B: Non‐ERM countries ∀ j exchange rate currencies ($) 47.4100 0.0 ∀ j exchange rate bonds (l.c.) 11.3007 0.04573 ∀ j exchange rate stocks (l.c.) 16.3413 0.00593 ρbji =0 ∀ j interest rate bonds 17.2587 0.00403 ρsji =0 ∀ j interest rate stocks 11.8358 0.03711 Note *Sample period 1977:11–1997:10. Computed using Newey‐‐West (1987) heteroscedasticity and autocorrelation consistent covariance matrix estimator. Open in new tab Table 5 Joint Tests for Pricing of Idiosyncratic Exchange Rate and Interest Rates Risks* Null hypothesis . Risk . Asset . Joint Wald test . p‐value . A: Core ERM countries exchange rate currencies ($) 12.7396 0.02594 exchange rate bonds (l.c.) 7.8618 0.16402 exchange rate stocks (l.c.) 2.2047 0.82016 ρbji =0 ∀ j interest rate bonds 11.7580 0.03826 ρsji =0 ∀ j interest rate stocks 8.3415 0.13839 B: Non‐ERM countries ∀ j exchange rate currencies ($) 47.4100 0.0 ∀ j exchange rate bonds (l.c.) 11.3007 0.04573 ∀ j exchange rate stocks (l.c.) 16.3413 0.00593 ρbji =0 ∀ j interest rate bonds 17.2587 0.00403 ρsji =0 ∀ j interest rate stocks 11.8358 0.03711 Null hypothesis . Risk . Asset . Joint Wald test . p‐value . A: Core ERM countries exchange rate currencies ($) 12.7396 0.02594 exchange rate bonds (l.c.) 7.8618 0.16402 exchange rate stocks (l.c.) 2.2047 0.82016 ρbji =0 ∀ j interest rate bonds 11.7580 0.03826 ρsji =0 ∀ j interest rate stocks 8.3415 0.13839 B: Non‐ERM countries ∀ j exchange rate currencies ($) 47.4100 0.0 ∀ j exchange rate bonds (l.c.) 11.3007 0.04573 ∀ j exchange rate stocks (l.c.) 16.3413 0.00593 ρbji =0 ∀ j interest rate bonds 17.2587 0.00403 ρsji =0 ∀ j interest rate stocks 11.8358 0.03711 Note *Sample period 1977:11–1997:10. Computed using Newey‐‐West (1987) heteroscedasticity and autocorrelation consistent covariance matrix estimator. Open in new tab We also find that idiosyncratic interest rate seems to be priced in bond markets (see row 4 of panels A and B), but in stock markets the evidence against the null is only significant for non‐ERM countries (see row 5). In this respect, it is important to bear in mind that all tests have, ceteris paribus, lower power in the stock return equations because the standard deviation of risks specific to stocks in each country are much higher (compare the idiosyncratic variances in Tables 2‐‐4). Our overall negative results are perhaps not totally surprising in view of the existence of exchange controls (at least for part of the sample) and other barriers to cross‐border investment, as well as informational asymmetries, behavioural biases, liquidity premia etc; (see also Section 2.4). Given that the individual coefficients show great dispersion, being even negative sometimes, and that the effects of exchange rate and interest rate volatility movements on risk premia may partly compensate each other, we have attempted to measure the net effects of idiosyncratic exchange rate and interest rate volatility on each asset by computing the differences in fitted values between alternative and null. Note that, in this way, each country acts as its own control. Figure 2a presents the average net effect on currency returns across ERM and non‐ERM countries, while Figures 2b and 2c display the analogous effects for bond and stock returns respectively. The corresponding sample means and relevant t‐statistics are reported in Table 6. Please note the changes in scale, which reflect the differences in average risk premia across asset classes. Table 6 Net Effect of Idiosyncratic Exchange Rate and Interest Volatility on Returns* Asset . Countries . Average . t‐statistic . p‐value . Currencies ($) Core ERM 0.029019 35.3667 0 Non‐ERM 0.022681 5.0649 0 Difference −0.006338 −1.4525 0.146 Bonds (l.c.) Core ERM −0.022311 −1.5830 0.113 Non‐ERM 0.049634 5.1728 0 Difference 0.071945 5.1430 0 Stocks (l.c.) Core ERM −0.020553 −2.5034 0.012 Non‐ERM 0.065158 2.6136 0.009 Difference 0.085711 3.4089 0.001 Asset . Countries . Average . t‐statistic . p‐value . Currencies ($) Core ERM 0.029019 35.3667 0 Non‐ERM 0.022681 5.0649 0 Difference −0.006338 −1.4525 0.146 Bonds (l.c.) Core ERM −0.022311 −1.5830 0.113 Non‐ERM 0.049634 5.1728 0 Difference 0.071945 5.1430 0 Stocks (l.c.) Core ERM −0.020553 −2.5034 0.012 Non‐ERM 0.065158 2.6136 0.009 Difference 0.085711 3.4089 0.001 Note *Sample period 1977:11–1997:10. Computed using Newey‐‐West (1987) heteroscedasticity and autocorrelation consistent covariance matrix estimator. Open in new tab Table 6 Net Effect of Idiosyncratic Exchange Rate and Interest Volatility on Returns* Asset . Countries . Average . t‐statistic . p‐value . Currencies ($) Core ERM 0.029019 35.3667 0 Non‐ERM 0.022681 5.0649 0 Difference −0.006338 −1.4525 0.146 Bonds (l.c.) Core ERM −0.022311 −1.5830 0.113 Non‐ERM 0.049634 5.1728 0 Difference 0.071945 5.1430 0 Stocks (l.c.) Core ERM −0.020553 −2.5034 0.012 Non‐ERM 0.065158 2.6136 0.009 Difference 0.085711 3.4089 0.001 Asset . Countries . Average . t‐statistic . p‐value . Currencies ($) Core ERM 0.029019 35.3667 0 Non‐ERM 0.022681 5.0649 0 Difference −0.006338 −1.4525 0.146 Bonds (l.c.) Core ERM −0.022311 −1.5830 0.113 Non‐ERM 0.049634 5.1728 0 Difference 0.071945 5.1430 0 Stocks (l.c.) Core ERM −0.020553 −2.5034 0.012 Non‐ERM 0.065158 2.6136 0.009 Difference 0.085711 3.4089 0.001 Note *Sample period 1977:11–1997:10. Computed using Newey‐‐West (1987) heteroscedasticity and autocorrelation consistent covariance matrix estimator. Open in new tab If we ignore the negative spikes, which correspond to devaluations in the Swedish krona, the evidence on currency returns suggests that, for both groups of countries, idiosyncratic exchange rate variability significantly increases short interest rate differentials for a given expected depreciation. This could explain the convergence of short rates in Europe during periods of low intra‐European exchange rate volatility. Therefore, a system that reduced both the expected level and the volatility of idiosyncratic exchange rate movements would seem to reduce the riskless component of the cost of capital. The evidence for bond and stock markets is less clear cut, with a small negative effect in the ERM bloc on average over time, and a significantly larger, positive effect in the other group. Nevertheless, a comparison of Figures 2b and c with Figure 1a suggests that periods in which local exchange rate volatility has risen have tended to be associated with increases in the required rate of return on bonds and equities. If with all the usual caveats in place, we were to extrapolate these results to a currency system such as EMU, in which country‐specific exchange rate uncertainty has almost vanished, and the interest rate policy of the member countries is not only common, but more importantly, released from the obligation of maintaining intra‐European exchange rate stability, our tentative conclusion would be that, ceteris paribus, the cost of capital in euroland is likely to be lower than it would have been otherwise, although the gains would probably be fairly small. 2.4. The Gains from Globalisation The international asset pricing model described in Section 1.1 implicitly assumes that European financial markets are integrated. Specifically, we have assumed that idiosyncratic risk is not priced, and that the price of risk associated with each of the underlying common factors is the same for all countries. Therefore, one way of testing for market integration is to examine if idiosyncratic factors are priced. We have already done this in the previous section for exchange rate and interest rate risks. The results of the tests for the pricing of country‐specific residual market risk in the equity markets show that the coefficients are not jointly significantly different from zero (Table 7, row 1), although again this is probably due to lack of power. Table 7 Additional Joint Tests for Market Integration* Null hypothesis . Risk . Asset . Joint Wald test . p‐value . ρsjm =0 ∀ j Residual market Stocks 13.1805 0.21376 Exchange rate Currencies ($) 16.1045 0.09668 Exchange rate Bonds (l.c.) 31.3994 0.00050 Exchange rate Stocks (l.c.) 103.3475 0.0 θbji =0 ∀ j Interest rate Bonds 11.2748 0.33651 θsji =0 ∀ j Interest rate Stocks 26.1813 0.00350 θsjm =0 ∀ j Residual market Stocks 142.2266 0.0 Null hypothesis . Risk . Asset . Joint Wald test . p‐value . ρsjm =0 ∀ j Residual market Stocks 13.1805 0.21376 Exchange rate Currencies ($) 16.1045 0.09668 Exchange rate Bonds (l.c.) 31.3994 0.00050 Exchange rate Stocks (l.c.) 103.3475 0.0 θbji =0 ∀ j Interest rate Bonds 11.2748 0.33651 θsji =0 ∀ j Interest rate Stocks 26.1813 0.00350 θsjm =0 ∀ j Residual market Stocks 142.2266 0.0 Note *Sample period 1977:11–1997:10. Computed using Newey‐‐West (1987) heteroscedasticity and autocorrelation consistent covariance matrix estimator. Open in new tab Table 7 Additional Joint Tests for Market Integration* Null hypothesis . Risk . Asset . Joint Wald test . p‐value . ρsjm =0 ∀ j Residual market Stocks 13.1805 0.21376 Exchange rate Currencies ($) 16.1045 0.09668 Exchange rate Bonds (l.c.) 31.3994 0.00050 Exchange rate Stocks (l.c.) 103.3475 0.0 θbji =0 ∀ j Interest rate Bonds 11.2748 0.33651 θsji =0 ∀ j Interest rate Stocks 26.1813 0.00350 θsjm =0 ∀ j Residual market Stocks 142.2266 0.0 Null hypothesis . Risk . Asset . Joint Wald test . p‐value . ρsjm =0 ∀ j Residual market Stocks 13.1805 0.21376 Exchange rate Currencies ($) 16.1045 0.09668 Exchange rate Bonds (l.c.) 31.3994 0.00050 Exchange rate Stocks (l.c.) 103.3475 0.0 θbji =0 ∀ j Interest rate Bonds 11.2748 0.33651 θsji =0 ∀ j Interest rate Stocks 26.1813 0.00350 θsjm =0 ∀ j Residual market Stocks 142.2266 0.0 Note *Sample period 1977:11–1997:10. Computed using Newey‐‐West (1987) heteroscedasticity and autocorrelation consistent covariance matrix estimator. Open in new tab The other way to assess if European markets are integrated is to test whether the prices of risk associated with the common factors are the same across countries in each of the three asset classes. The null hypothesis of common exchange rate risk evaluation is not rejected in the currency markets (Table 7, row 2), but it is strongly rejected in the bond and stock markets (rows 3 and 4). As for the common interest rate factor, the assumption of common valuation is not rejected in European bond markets (see row 5), but it is rejected in stock markets (row 6). Finally, we also find significant differential pricing for common residual market risks (see row 7). One attractive reinterpretation of our lack of integration results is to say that international investors are incompletely diversified, which is consistent with the well known fact that most investors show a home bias, in the sense that they hold only a relatively small proportion of their assets outside their country; see, for example, Kang and Stulz (1997) for a discussion and references. In this respect, an important indirect effect of EMU would be the elimination of some of the elements that limit cross‐border investments in Europe, such as the restrictions on foreign asset holdings by pension funds, insurance companies and other financial institutions, or the behavioural biases traditionally shown by individual investors in favour of assets denominated in their own domestic currency. In fact, there is ample anecdotal evidence that such a process is already taking place at a very rapid rate. To gauge the potential gains from increased market integration, we are going to follow Stulz (1999) in comparing stock market risk premia under full integration with the risk premia that would prevail in the context of fully segmented markets. In this sense, note that, although the distinction between common and specific risk becomes irrelevant in the latter context, it is still possible to write the unanticipated components of stock returns in terms of three orthogonal sources of risk: (10) Since the stock portfolio for each country corresponds to a diversified basket of domestic stocks, an argument similar to the one presented in Section 1.1 would then result in a domestic version of the APT (11) where ϕje, ϕji and ϕjm are the prices of risk in country j. Note that (11) is a special case of our general alternative hypothesis (5), with the restrictions ρsjk=ϕsjk=τk and θsjk=ϕsjkβjsk(βjsk−1) for k=e,i,m. Assuming as in Stulz (1999) that financial market integration does not affect the prices of residual market risk, so that ϕjm=τm, we can assess whether there would be gains (on average over time) to each country from stock market integration by comparing these quantities:14 Table 8 reports the difference between the right‐ and left‐hand sides of the above expression for the European countries in our database. Apart from the fact that all the differences are highly significantly positive, the other striking result is that there is substantial variation across countries. In particular, the largest average gains correspond to countries such Italy, Sweden or Spain, which tend to have both large βsjms , and especially, very large idiosyncratic variances. As a result, those countries are the ones that a priori would benefit most from an increase in stock market integration. In contrast, the benefits for countries such as the Netherlands, Switzerland or the UK, whose stock markets have significantly smaller residual market risk, probably because they already have closer links with world markets, would be smaller. If we multiply those differences by 0.03, which is roughly our implicitly estimate of τm (see footnote 10), our results suggest that the potential gains from stock market integration could be rather large. Nevertheless, it is important to emphasise that those gains should only be taken as indicative, in view of the large standard error associated with the price of risk coefficient, and the fact that we are comparing a situation of full segmentation with another of complete integration. In practice, of course, markets are neither fully segmented, nor fully integrated and, moreover, the transition from one state to the other is typically a gradual process, whose effects are partly anticipated by investors. In this respect, Hardouvelis et al. (2000) find that the degree of integration of European stock markets increased substantially after 1995, when forward interest differentials vis‐ à‐vis Germany began to narrow in anticipation of EMU membership. Table 8 Average Gains from Increased Stock Market Integration (local currency) * Country . . βsjmλm . Difference . Belgium 19.6857 9.90221 9.78350 (4.43808) (3.42981) (2.30608) Denmark 18.6341 7.31373 11.3204 (2.06554) (1.74589) (2.02143) France 30.5656 13.5519 17.0136 (5.52645) (3.74877) (2.84319) Germany 25.5600 12.2155 13.3445 (5.68106) (4.08642) (2.40860) Italy 45.7898 13.1742 32.6156 (6.33277) (2.51710) (5.46835) Netherlands 17.7324 11.1545 6.57790 (3.73677) (3.32912) (1.21043) Spain 32.5438 12.3362 20.2077 (5.89328) (3.97726) (3.08184) Sweden 37.2293 13.5793 23.6500 (5.21221) (3.48773) (3.14164) Switzerland 16.7930 10.1943 6.59876 (4.33630) (3.78708) (1.06078) UK 20.0163 11.6095 8.40678 (5.49114) (4.02881) (1.70797) Country . . βsjmλm . Difference . Belgium 19.6857 9.90221 9.78350 (4.43808) (3.42981) (2.30608) Denmark 18.6341 7.31373 11.3204 (2.06554) (1.74589) (2.02143) France 30.5656 13.5519 17.0136 (5.52645) (3.74877) (2.84319) Germany 25.5600 12.2155 13.3445 (5.68106) (4.08642) (2.40860) Italy 45.7898 13.1742 32.6156 (6.33277) (2.51710) (5.46835) Netherlands 17.7324 11.1545 6.57790 (3.73677) (3.32912) (1.21043) Spain 32.5438 12.3362 20.2077 (5.89328) (3.97726) (3.08184) Sweden 37.2293 13.5793 23.6500 (5.21221) (3.48773) (3.14164) Switzerland 16.7930 10.1943 6.59876 (4.33630) (3.78708) (1.06078) UK 20.0163 11.6095 8.40678 (5.49114) (4.02881) (1.70797) Note * Sample period 1977:11–1997:10. Newey‐‐West (1987) heteroscedasticity and autocorrelation robust standard errors in parenthesis. Open in new tab Table 8 Average Gains from Increased Stock Market Integration (local currency) * Country . . βsjmλm . Difference . Belgium 19.6857 9.90221 9.78350 (4.43808) (3.42981) (2.30608) Denmark 18.6341 7.31373 11.3204 (2.06554) (1.74589) (2.02143) France 30.5656 13.5519 17.0136 (5.52645) (3.74877) (2.84319) Germany 25.5600 12.2155 13.3445 (5.68106) (4.08642) (2.40860) Italy 45.7898 13.1742 32.6156 (6.33277) (2.51710) (5.46835) Netherlands 17.7324 11.1545 6.57790 (3.73677) (3.32912) (1.21043) Spain 32.5438 12.3362 20.2077 (5.89328) (3.97726) (3.08184) Sweden 37.2293 13.5793 23.6500 (5.21221) (3.48773) (3.14164) Switzerland 16.7930 10.1943 6.59876 (4.33630) (3.78708) (1.06078) UK 20.0163 11.6095 8.40678 (5.49114) (4.02881) (1.70797) Country . . βsjmλm . Difference . Belgium 19.6857 9.90221 9.78350 (4.43808) (3.42981) (2.30608) Denmark 18.6341 7.31373 11.3204 (2.06554) (1.74589) (2.02143) France 30.5656 13.5519 17.0136 (5.52645) (3.74877) (2.84319) Germany 25.5600 12.2155 13.3445 (5.68106) (4.08642) (2.40860) Italy 45.7898 13.1742 32.6156 (6.33277) (2.51710) (5.46835) Netherlands 17.7324 11.1545 6.57790 (3.73677) (3.32912) (1.21043) Spain 32.5438 12.3362 20.2077 (5.89328) (3.97726) (3.08184) Sweden 37.2293 13.5793 23.6500 (5.21221) (3.48773) (3.14164) Switzerland 16.7930 10.1943 6.59876 (4.33630) (3.78708) (1.06078) UK 20.0163 11.6095 8.40678 (5.49114) (4.02881) (1.70797) Note * Sample period 1977:11–1997:10. Newey‐‐West (1987) heteroscedasticity and autocorrelation robust standard errors in parenthesis. Open in new tab 3. Conclusions In this paper, we use monthly data on currency, bond and stock returns for ten European countries over the period 1977–97 to estimate a dynamic multi‐factor APT model with time‐varying volatility in both common and idiosyncratic factors. From the methodological point of view, our main contribution is to exploit the cross‐sectional dimension of the model to construct diversified world portfolios which, when added to the original list of assets, can be used to develop simple, but nevertheless consistent GMM‐based methods of estimation and testing. We find that, controlling for exchange rate variations, world bond and stock returns are positively correlated. We also find that the average expected returns on the common risks are all positive, but not very precisely estimated. Moreover, our results show that increases in the common exchange rate, interest rate and residual market factors are associated with higher European currency, bond and stock returns respectively. They also suggest that a local currency appreciation generally leads to higher (local currency) bond returns, while (local currency) stock returns in Europe suffer when the dollar depreciates. In addition, we find that controlling for exchange rate variations, the idiosyncratic terms for bonds and stocks are positively and significantly correlated within countries. Furthermore, our findings indicate that a target zone system such as the EMS does reduce exchange rate volatility as long as it remains credible, and that average interest rate volatility has actually been smaller in ERM countries than in non‐ERM ones. At the same time, our results confirm that reductions in idiosyncratic exchange rate volatility may sometimes be achieved at the expense of increases in local interest rate volatility. Importantly, our evidence also suggests that a system that reduces both the expected level and the volatility of idiosyncratic exchange rate movements is likely to reduce the riskless component of the cost of capital. The evidence for bond and stock markets is less clear cut, although periods in which local exchange rate volatility has risen have tended to be associated with increases in the required rate of return on bonds and equities. Nevertheless, the effects that we uncover are small. We also find overall negative results on market integration. For that reason, and given that an important indirect effect of EMU would be an increase in the degree of integration of European financial markets, we compare stock market risk premia under full integration with the risk premia that would prevail in the context of fully segmented markets. Our results suggest that such an upper bound on the potential gains from stock market integration could be rather large. Finally, it is worth bearing in mind that there are other important channels through which the removal of financial market segmentation by means of regional arrangements such as the EMS could lead to a decrease in funding costs. Given the separation of ownership and control that exists in the quoted corporate sector, the most important mechanism will be the reduction in the agency costs faced by firms raising outside capital, which are not directly related to the required rate of return used in present discounted value calculations; see Stulz (1999). At the same time, the creation of a wider, deeper and more liquid capital market is likely to intensify the pressure on European firms to pay more attention to shareholders, and to seek to raise the value of their stocks. In this respect, cross‐border mergers and acquisitions show a clear upward trend. In addition, increased competition by banks and other financial institutions may also result in a reduction in the cost of capital for smaller unquoted companies, which account for a significant proportion of European investment. Therefore, there is little doubt that the measuring of the effects of ‘globalisation’ on the cost of capital would continue to be an area of active research. Footnotes 1 " See Ando and Auerbach (1988) for a discussion of the difficulties induced by differential taxation in cross‐country comparisons of the debt component of the cost of capital. 2 " Strictly speaking, excess returns should be defined in terms of arithmetic returns, rather than geometric returns. Nevertheless, the approximation error is usually irrelevant in empirical work; see McCulloch (1975) . 3 " See Sentana and Fiorentini (2001) for the implications of this assumption on the identification of the factors and their loadings. 4 " Their definition is asymptotic, and amounts in our case to the largest eigenvalue of the (conditional) idiosyncratic covariance matrix remaining uniformly bounded as the number of countries , N , goes to infinity. 5 " If the stochastic discount factor ξt contains some idiosyncratic risk, the APT expression (4) becomes approximate, as in Ross’ (1976) original formulation, and it is only possible to prove that the pricing errors would be negligible on average cross‐sectionally, but not necessarily so for each asset. Connor (1984) provides conditions that guarantee exact asset pricing in a competitive equilibrium set‐up. 6 " Ross’ (1976) results were implicitly derived for a closed economy. However, if exchange rates belong to the set of factors, then the APT can be readily generalised to an international setting; see Solnik (1983) and Ikeda (1991) . 7 " Maximum likelihood estimates of the trivariate 3‐factor GARCH in mean model for in (9) can be easily obtained by replacing OLS with univariate GARCH‐M regressions in (a), (b) and (c) above, but the standard errors usually computed would be inconsistent due to the generated regressor problem. 8 " Maximum likelihood estimates of the factor loadings for each country model in (8) can be obtained by replacing OLS in (d), (e) and (f) with univariate GARCH regressions, but again the standard errors usually computed would be inconsistent because of the generated regressor problem. 9 " It is in principle possible, albeit non‐trivial, to estimate the optimal weightings of the basis portfolios together with all the other parameters; (see for example KSW). In any case, note that the consistency of the estimation methods described in Section 1.3 would not be affected if the international portfolios that we construct were not fully diversified, although it would be necessary to develop an asymptotic theory in which both N and T increase at possibly the same rate in order to robustify our inference. Both issues are left for further research. 10 " A complete list of ERM realignments between March 1979 and December 1994 can be found in Ayuso and Pérez‐Jurado (1997) . 11 " A useful way to measure the ‘distance’ between ML and GMM parameter estimates is to look at the differences between the implied basis portfolios for common and specific factors. In this respect, we find that both sets of estimates yield very similar results, with correlations ranging from 0.979 to 1, with an average value of 0.997. 12 " If we compute ‘average’ price of risk coefficients for k = e , i , m as , we obtain 0.001177, 0.140906 and 0.030949, which although strictly speaking inconsistent, are rather similar to the corresponding ML estimates (0.001627, 0.121238 and 0.028841). 13 " As discussed in Section 1.3, our estimation procedure is based on the maintained assumption that the covariance matrix of the idiosyncratic terms satisfies the approximate zero factor structure of Chamberlain and Rothschild (1983). We have informally assessed whether such an assumption is correct on the basis of the correlation matrix of the estimated idiosyncratic terms for all the countries in our dataset. In this respect, we find that the average of its off‐diagonal elements (squared) is 0.017518. Similarly, we find that its largest eigenvalue only explains 10.5% of its trace, as opposed to 13% when we only consider the European countries. 14 " Given the definition of beta as the ratio of covariance to variance, and the fact that for innovations the unconditional variance coincides with the average of the conditional one, we can equivalently write this expression in terms of covariances, as in Stulz (1999) , yielding * " This is a thoroughly revised version of Sentana et al. (1995) , which extended the analysis of stock returns in Sentana et al. (1992) to currencies and bonds. I am very grateful to Mushtaq Shah and Sushil Wadhwani for their contributions to the earlier versions of this paper. I would also like to thank Manuel Arellano, Juan Ayuso, Philippe Bacchetta, José M. Campa, Greg Connor, Charles Goodhart, Mike Wickens and participants at seminars at AUEB (Athens), Bank of England, Bank of Spain, the Board of Governors of the Federal Reserve, Fundación BBV (Bilbao), IAE (Barcelona), LSE, UIMP (Santander), and the 1995 Royal Economic Society Conference for their comments. The editor and three anonymous referees have also helped me greatly to improve the paper. Of course, the usual caveat applies. Thanks are also due to Juan Ayuso, Martin Brookes, Aelred Conelly, Simon Cook, José A. Cuenca and Andrew Haldane for their help in collecting the data and to Isabel Figuerola‐Ferretti and Angel León for research assistance. This work was initiated at the LSE Financial Markets Group, whose financial support for the first version as part of the ESRC project ‘The Efficiency and Regulation of Financial Markets’ is gratefully acknowledged. References Ando , A. and Auerbach , A. J. ( 1988 ). ‘The cost of capital in the United States and Japan: a comparison’ , Journal of the Japanese and International Economy , vol. 2 , pp. 134 – 58 . Google Scholar Crossref Search ADS WorldCat Ayuso , J. and Pérez‐Jurado , M. ( 1997 ). ‘Devaluations and depreciation expectations in the EMS’ , Applied Economics , vol. 29 , pp. 471 – 84 . Google Scholar Crossref Search ADS WorldCat Batchelor , R. ( 1985 ). Evidence to the House of Lords Select Committee. Bekaert , G. and Hodrick , R. J. ( 1992 ). ‘Characterizing predictable components in excess returns on equity and foreign exchange markets’ , Journal of Finance , vol. 47 , pp. 467 – 509 . Google Scholar Crossref Search ADS WorldCat Bollerslev , T. and Wooldridge , J. M. ( 1992 ). ‘Quasi‐maximum likelihood estimation and inference in dynamic models with time‐varying variances’ , Econometric Reviews , vol. 11 , pp. 143 – 72 . Google Scholar Crossref Search ADS WorldCat Cochrane , J. H. ( 2001 ). Asset Pricing . Princeton: Princeton University Press . Google Scholar Google Preview OpenURL Placeholder Text WorldCat COPAC Connor , G. ( 1984 ). ‘A unified beta pricing theory’ , Journal of Economic Theory , vol. 34 , pp. 13 – 31 . Google Scholar Crossref Search ADS WorldCat Connor , G. and Korajczyk , R. A. ( 1988 ). ‘Risk and return in an equilibrium APT: an application of a new test methodology’ , Journal of Financial Economics , vol. 21 , pp. 255 – 89 . Google Scholar Crossref Search ADS WorldCat European Commission ( 1990 ). ‘One market, one money’ , European Economy special issue , October. Google Scholar Google Preview OpenURL Placeholder Text WorldCat COPAC Forni , M. and Reichlin , L. ( 1998 ). ‘Let's get real: a factor analytical approach to disaggregated business cycle dynamics’ , Review of Economic Studies vol. 65 , pp. 453 – 73 . Google Scholar Crossref Search ADS WorldCat Hardouvelis , G. , Malliaropulos , D. and Priestley , R. ( 2000 ). ‘EMU and European stock market integration’ , mimeo, Norwegian School of Management. Ikeda , S. ( 1991 ). ‘Arbitrage asset pricing under exchange risk’ , Journal of Finance , vol. 46 , pp. 447 – 55 . Google Scholar Crossref Search ADS WorldCat Jarrow , R. , Maximovich , M. and Ziemba , W. ( 1995 ). Handbook of Modern Finance , Amsterdam: North Holland. Google Scholar Google Preview OpenURL Placeholder Text WorldCat COPAC Kang , J. ‐K. and Stulz , R. ( 1997 ). ‘Why is there a home bias? An analysis of foreign portfolio equity ownership in Japan’ , Journal of Financial Economics , vol. 46 , pp. 3 – 28 . Google Scholar Crossref Search ADS WorldCat King , M. , Sentana , E. and Wadhwani , S. ( 1994 ). ‘Volatility and links between national stock markets’ , Econometrica , vol. 62 , pp. 901 – 33 . Google Scholar Crossref Search ADS WorldCat McCulloch , J. H. ( 1975 ). ‘Operational aspects of the Siegel paradox’ , Quarterly Journal of Economics , vol. 89 , pp. 170 – 2 . Google Scholar Crossref Search ADS WorldCat Newey , W. and West , K. D. ( 1987 ). ‘A simple positive semi‐definite heteroskedasticity and autocorrelation consistent covariance matrix’ , Econometrica , vol. 55 , pp. 703 – 6 . Google Scholar Crossref Search ADS WorldCat Pagan , A. and Ullah , A. ( 1988 ). ‘The econometric analysis of models with risk terms’ , Journal of Applied Econometrics , vol. 3 , pp. 87 – 105 . Google Scholar Crossref Search ADS WorldCat Ross , S. A. ( 1976 ). ‘The arbitrage theory of capital asset pricing’ , Journal of Economic Theory , vol. 13 , pp. 641 – 60 . Google Scholar Crossref Search ADS WorldCat Sentana , E. ( 1997a ). ‘Risk and return in the Spanish stock market: some evidence from individual assets’ , Investigaciones Económicas , vol. 21 , pp. 297 – 359 . OpenURL Placeholder Text WorldCat Sentana , E. ( 1997b ). ‘Estimation of a triangular seemingly unrelated regression system by OLS’ , Econometric Theory , vol. 13 , Problem 97.2.2. OpenURL Placeholder Text WorldCat Sentana , E. ( 2002 ). ‘Factor representing portfolios in large asset markets’ , forthcoming in the Journal of Econometrics . OpenURL Placeholder Text WorldCat Sentana , E. and Fiorentini , F. ( 2001 ). ‘Identification, estimation and testing of conditionally heteroskedastic factor models’ , Journal of Econometrics , vol. 102 , pp. 143 – 64 . Google Scholar Crossref Search ADS WorldCat Sentana , E. , Shah , M. and Wadhwani , S. ( 1992 ). ‘Has the EMS reduced the cost of capital?’ , LSE FMG Discussion Paper 134 . Sentana , E. , Shah , M. and Wadhwani , S. ( 1995 ). ‘Has the EMS reduced the cost of capital?’ , CEMFI Working Paper 9514 . Shiller , R. J. ( 1979 ). ‘The volatility of long‐term interest rates and expectations models of the term structure’ , Journal of Political Economy , vol. 87 , pp. 1190 – 219 . Google Scholar Crossref Search ADS WorldCat Solnik , B. ( 1983 ). ‘International arbitrage pricing theory’ , Journal of Finance , vol. 18 , pp. 449 – 57 . Google Scholar Crossref Search ADS WorldCat Stock , J. H. and Watson , M. W. ( 1998 ). ‘Diffusion indexes’ , NBER Working Paper 6702 . Stulz , R. ( 1995 ). ‘International asset pricing: an integrative survey’ , in Jarrow et al . (1995), pp. 201 – 23 . Stulz , R. ( 1999 ). ‘Globalization of equity markets and the cost of capital’ , NBER Working Paper 7021 . Appendix Data Appendix Details of the data series used are as follows (where appropriate the name of the series is followed by its Datastream code). Stock prices and dividend yields: *Morgan Stanley Capital International Perspectives Exchange rates: *US$‐Aus$, end of period (AUOCEXCH) *US$‐BelFr, end of period (BGOCEXCH) *US$‐CAN$, end of period (CNOCEXCH) *US$‐DENKr, end of period (DKOCEXCH) *US$‐FraFr, end of period (FROCEXCH) *US$‐DM, end of period (BDOCEXCH) *US$‐ItaLit, end of period (ITOCEXCH) *US$‐Yen, end of period (JPOCEXCH) *US$‐NetFl, end of period (NLOCEXCH) *US$‐SpPta, end of period (ESOCEXCH) *US$‐SweKr, end of period (SDOCEXCH) *US$‐SwiFr, end of period (SWOCEXCH) *US$‐Stg, end of period (UKOCEXCH) Short interest rates: *Belgium, Canada, Denmark, France, Germany, Italy, Japan, Netherlands, Sweden, Switzerland, UK, US: 1‐month Eurocurrency rate from the Bank of England. *Australia: Yield on 3‐month Treasury Bill (AUTRSBL%) *Spain: pre 1983:1: 3‐month interbank rate from the Bank of Spain; post 1983:1: 1‐month Eurocurrency rate. Bond returns: *Canada, France, Germany, Japan, Netherlands, Switzerland, UK, US: pre 1978:1: Shiller (1979) approximation on 10‐year bond yields from Goldman Sachs; post 1978:2: Salomon Brothers bond return indices. *Australia: pre‐1984:10: Shiller (1979) approximation on 10‐year bond yield; post 1984:11: Salomon Brothers bond return indices *Italy: pre 1990:12: Shiller (1979) approximation on 10 year bond yields from Goldman Sachs; post 1991:1: Salomon Brothers bond return indices. *Belgium, Denmark, Sweden: pre 1985:3: Shiller (1979) approximation on 10 year bond yields from Goldman Sachs; 1985:4‐‐1990:12: Datastream bond return indices (Belgium (ABGGYG4(RI)), Denmark (ADKGYG4(RI)), Sweden (ASDGYG4(RI)); post 1991:1: Salomon Brothers bond return indices. *Spain: pre 1982:1: Shiller (1979) approximation on bond yields from Bank of Spain; 1982:2‐1985:3: Shiller (1979) approximation on bond yields from Goldman Sachs; 1985:3:1990:12: Datastream bond return indices (AESGYG4(RI)); post 1991:1: Salomon Brothers bond return indices. © Royal Economic Society 2002 http://www.deepdyve.com/assets/images/DeepDyve-Logo-lg.png The Economic Journal Oxford University Press http://www.deepdyve.com/lp/oxford-university-press/did-the-ems-reduce-the-cost-of-capital-JabaIFh7Q5

References (29)

Gregory Connor, Robert Korajczyk (1988)
Risk and Return in an Equilibrium Apt: Application of a New Test Methodology
T. Bollerslev, J. Wooldridge (1992)
Quasi-maximum likelihood estimation and inference in dynamic models with time-varying covariances
Econometric Reviews, 11
Juan Ayuso, M. Perez-Jurado (1997)
Devaluations and depreciation expectations in the EMS
Applied Economics, 29
G. Hardouvelis, R. Priestley, D. Malliaropulos (2001)
Emu and European Stock Market Integration
Capital Markets eJournal
Shinsuke Ikeda (1991)
Arbitrage Asset Pricing under Exchange Risk
Journal of Finance, 46
René Stulz (1999)
Globalization of Equity Markets and the Cost of Capital
Corporate Finance: Capital Structure & Payout Policies
A. Ando, A. Auerbach (1988)
The cost of capital in the United States and Japan: A comparison
Journal of The Japanese and International Economies, 2
Shiller (1979) approximation on 10-year bond yield; post 1984:11: Salomon Brothers bond return indices *Italy: pre 1990:12: Shiller (1979) approximation on 10 year bond yields from Goldman Sachs
G. Fiorentini, Enrique Sentana (2001)
Identification, Estimation and Testing of Conditionally Heteroskedastic Factor Models
Econometrics eJournal
R. Shiller (1979)
The Volatility of Long-Term Interest Rates and Expectations Models of the Term Structure
Journal of Political Economy, 87
Bruno Solnik (1983)
International Arbitrage Pricing Theory
Journal of Finance, 38
René Stulz (1994)
International Portfolio Choice and Asset Pricing: An Integrative Survey
NBER Working Paper Series
M. King, Enrique Sentana, S. Wadhwani (1990)
Volatiltiy and Links between National Stock Markets
Capital Markets: Asset Pricing & Valuation eJournal
J. Mcculloch (1975)
Operational Aspects of the Siegel Paradox
Quarterly Journal of Economics, 89
Enrique Sentana (1997)
Risk and return in the Spanish stock market: some evidence from individual assets
Investigacion Economica, 21
S. Ross (1976)
The arbitrage theory of capital asset pricing
Journal of Economic Theory, 13
Jun-Koo Kang, René Stulz (1995)
Why is There a Home Bias? An Analysis of Foreign Portfolio Equity Ownership in Japan
Capital Markets: Asset Pricing & Valuation
G. Bekaert, R. Hodrick (1991)
Characterizing Predictable Components in Excess Returns on Equity and Foreign Exchange Markets
Capital Markets: Asset Pricing & Valuation
M. Wolf (1996)
One Market, One Money
Enrique Sentana, M. Shah, S. Wadhwani (1995)
Has the EMS Reduced the Cost of Capital
Gregory Connor (1984)
A Unified Beta Pricing Theory
Journal of Economic Theory, 34
Enrique Sentana (2000)
Factor Representing Portfolios in Large Asset Markets
Journal of Financial Abstracts eJournal
Shiller (1979) approximation on bond yields from Goldman Sachs; 1985:3:1990:12: Datastream bond return indices (AESGYG4(RI))
(1991)
ration’, mimeo, Norwegian School of Management
A. Pagan, A. Ullah (1988)
The Econometric Analysis of Models with Risk Terms
Journal of Applied Econometrics, 3
Mario Forni, L. Reichlin (1998)
Let's Get Real: A Factor Analytical Approach to Disaggregated Business Cycle Dynamics
The Review of Economic Studies, 65
Enrique Sentana (1997)
Estimation of a Triangular, Seemingly Unrelated, Regression System by OLS
Econometric Theory, 13
Whitney Newey, K. West (1986)
A Simple, Positive Semi-Definite, Heteroskedasticity and Autocorrelationconsistent Covariance Matrix
Econometrics eJournal
D. Logue (1984)
Handbook of Modern Finance

Publisher: Oxford University Press
Copyright: © Royal Economic Society 2002
ISSN: 0013-0133
eISSN: 1468-0297
DOI: 10.1111/1468-0297.00069
Publisher site: See Article on Publisher Site

Abstract

Abstract We propose a dynamic arbitrage pricing theory (APT) multi‐factor model with time‐varying volatility for currency, bond and stock returns for ten European countries over 1977–97. We exploit the cross‐sectional dimension of the model to construct world portfolios, which when added to the original list of assets, allow us to develop simple consistent methods of estimation and testing. Our results reject the implicit asset pricing restrictions, and suggest decreases in idiosyncratic exchange rate risk tend to lower the cost of capital, although the effect is small. We assess the potential gains from increased stock market integration. The European Commission argued (1990, p. 82) that ‘potentially the most important source of gains from European Monetary Union comes from the reduction in overall uncertainty EMU might provide’. They went on to contend that a reduction in exchange rate risk would reduce the risk premium, and that this reduction in the cost of capital would stimulate corporate investment. Nevertheless, as the report acknowledged, it is not a priori obvious that intra‐European exchange rate risk should affect the cost of capital. For example, firms might be able to hedge their exchange rate exposure through a variety of financial instruments. Similarly, those who diversify their investments globally may not be affected by idiosyncratic variations in a country's exchange rate. Therefore, a preliminary question that has to answered is whether country‐specific exchange rate factors are priced in the context of asset pricing models applied at the international level; see Stulz (1995) for a recent survey. Although those risks should not be rewarded in a world with complete market integration, the existence of capital controls or other legal impediments to cross‐border investment (such as limitations on the holdings of foreign securities by pension funds and insurance companies), informational asymmetries, illiquid markets, behavioural biases etc. suggest that idiosyncratic exchange rate risk is likely to be priced. Hence, whether or not EMU would reduce the cost of capital, and if so, by how much, is, ultimately, an empirical question. We attempt to throw some light on this issue within the framework of the dynamic version of the arbitrage pricing theory (APT) developed in King et al. (1994)(KSW hereinafter). Specifically, we use monthly data on currency, bond and stock returns for ten European countries over the period 1977–97 to estimate a multi‐factor model with time‐varying volatility in the underlying factors, in which the idiosyncratic components of returns are (almost) uncorrelated across countries, but their correlation structure is arbitrary within each country. From the methodological point of view, we exploit the cross‐sectional dimension of the model to construct diversified portfolios of European and non‐European securities, which when added to the original list of assets, allow us to develop simple consistent methods of estimation and testing. In particular, we employ a GMM estimation procedure that would be efficient under conditional homoscedasticity, but which remains consistent under the more realistic assumption of dynamic heteroscedasticity. Under the null, our model implies that country‐specific risks should not be priced. However as mentioned before, it is likely that these idiosyncratic risk components do indeed affect the required rate of return on bonds and stocks. If country‐specific exchange rate volatility is associated with higher stock returns, then systems that attempt to reduce nominal exchange rate variability, such as the Exchange Rate Mechanism (ERM) of the European Monetary System (EMS), may well reduce capital costs for firms that raise funds by issuing equity. Since the stock market is not the primary source of finance in some of the countries that we look at, Germany being the prime example, we also look at bond returns.1 In either case, though, we would still be looking at only one component of the cost of capital. Since we include currency returns in our set of assets, we can also say something about the effect of the EMS on the riskless interest rate component. However, under a target zone system such as the ERM, the cost of capital would not necessarily fall; for example, some authors such as Batchelor (1985) have argued that a credible system will inevitably increase interest rate volatility since the authorities are forced to defend the currency. If interest rate volatility is also positively associated with risk premia on bond and equity markets, and if ERM membership raised interest rate volatility, then it is theoretically possible that the EMS might even have increased the cost of capital. Again, this is an issue that we shall investigate. More generally, testing the cross‐equation asset pricing restrictions of our basic model enables us to examine the important question of whether European capital markets are integrated. In this respect, it is worth noting that an important indirect effect of EMU, in conjunction with the development of the single market for financial services, should be the elimination of many of the remaining barriers to cross‐border investments in the EU. To gauge the potential gains from increased market integration, we follow Stulz (1999), and compare stock market risk premia under full integration with the risk premia that would prevail in the context of completely segmented markets. The rest of the paper is organised as follows. Section 1 discusses our basic model and its estimation procedure. Section 2 reports the empirical results, and Section 3 contains our conclusions. This is followed by a Data Appendix. 1. Theory and Estimation 1.1. Asset Pricing Model We base our analysis in a world with a large number of countries, and assume that, for each country, there are three representative assets available: a 1‐period local currency deposit with safe gross return , a long‐term default‐free bond portfolio, whose random gross holding return over period t in local currency is , and a stock portfolio, whose random gross holding return in local currency is . Let be the spot exchange rate for country j at the end of period t in terms of the numeraire currency (US $ in our case), and let be the gross return on the safe asset for the US during period t in US$. The excess returns of the three representative assets for each country in terms of the numeraire currency will be given by (1) where and are the (continuously compounded) excess returns for bonds and stocks in local currency.2 Note that, under covered interest parity, is the variation in exchange rates corrected for the forward premium. Let and be the risk premia on currency, bond and stock returns for country j, and let and be the corresponding unanticipated (as of time t−1) components of returns. Given that simply reflects the unexpected variations in exchange rates against the common numeraire currency, and these are so highly correlated across countries, a single‐factor model should provide a realistic representation of their covariance structure. In this respect, note that full European monetary union would imply that would be the same for all member countries. On the other hand, while both bond and stock returns are additionally affected by interest rate movements, it is generally accepted that stock returns are also exposed to other risks. On this basis, we assume the following conditional factor structure for the innovations in returns: (2) or, in matrix notation, , where fet and fit are common factors representing systematic exchange rate and interest rate risks, fmt is a common residual market risk factor, vjet and vjit are exchange and interest rate risks specific to country j, vjmt represents other risks specific to country j stocks, and the βs and δs are the associated factor loadings that measure the sensitivity of the assets to the different factors, which we assume time‐invariant for any given unconditional normalisation of the factors. To guarantee that the are innovations, we assume that common and specific factors are unpredictable on the basis of past information. Following KSW, we also make the assumption that the common factors are conditionally orthogonal to each other, but allow them to have time‐varying conditional variances λet,λit and λmt.3 As for the idiosyncratic terms, which by definition are conditionally orthogonal to ft, we also assume that they are conditionally orthogonal to one another for a given j, with time‐varying conditional variances ωjet,ωjit and ωjmt. But note that since , the idiosyncratic unconditional covariance matrix remains totally unrestricted within a country. Finally, we assume that the idiosyncratic conditional covariance matrix has the approximate zero‐factor structure introduced by Chamberlain and Rothschild (1983), in which vjet,vjit and vjmt may be correlated across countries, but only mildly so in order to guarantee that full diversification applies.4 Under a mild no arbitrage condition, it is possible to prove that there is a stochastic discount factor, ξt, which prices the available assets by discounting their uncertain pay‐offs across different states of the world; see, for example Cochrane (2001). In terms of excess returns, in particular, we will have that (3) If (and only if) the stochastic discount factor contains no idiosyncratic risk, ξt will be correlated with , and only through ft, and the assets’ risk premia can be written as an exact linear combination of the volatility of the common factors, with weights proportional to the corresponding factor loadings. Specifically, (4) where τk (k=e,i,m) are the prices of risk corresponding to each factor, ie the amount of expected return that agents are willing to give away to reduce its variability by one unit. An alternative way of interpreting the above relationship can be obtained by noticing that risk premia are also linear combinations of the asset factor loadings or betas, with weights that are common to all assets. These weights can be understood as the risk premia of three factor mimicking portfolios, ie three unit‐cost, diversified portfolios of risky assets with unit loadings on the common exchange rate, interest rate and residual stock market risk factors respectively, and zero loadings on the others. In this respect, our model coincides with a conditional version of the exact5 APT pricing relationship.6 Importantly, since we explicitly consider currency returns, our model will also hold for local currency excess returns on bonds and stocks, which can be understood as fully hedged returns; see, for example, Bekaert and Hodrick (1992). Furthermore, it will also hold for stock and bond returns measured in a common (diversified) basket currency. Note that, if our asset pricing model is correct, risk prices depend on the factors, not on the assets, since otherwise there would be arbitrage opportunities. Furthermore, the model also implies that country‐specific risks should not be rewarded because they can be diversified away. As we will see in the next section, these fundamental restrictions are the basis of our tests. 1.2. Alternative Hypotheses To assess whether EMU would reduce the cost of capital and, if so, by how much, our most general alternative is given by (5) where and are defined in (4). This equation enables us to test various hypotheses of interest: We may ask if idiosyncratic European exchange rate variability is not priced in bond and stock markets, ie if ρbje = ρsje =0 ∀ j . If so, then it is difficult to see how the EMS could have affected the equity and bond components of the cost of capital by affecting exchange rate volatility. Similarly, testing if ρcje =0 ∀ j is also very interesting, as it would throw some light on the question of whether the observed convergence of short rates in Europe is explicable by the reduction in exchange rate volatility. In both cases, the values of ρaje ( a = c , b , s ) will allow us to measure the effects of eliminating such idiosyncratic risks. If idiosyncratic interest rate volatility is priced, ie ρbji ≠0 or ρsji ≠0, then the EMS might have affected the cost of capital, if it also affected interest rate volatility. It is possible that other sources of idiosyncratic risk are priced in the stock market, ie ρsjm ≠0, which would suggest that European stock markets are not fully integrated. We may also test for European capital market integration in the sense of asking whether the different prices of risk are common across countries. To do so, we can ask whether freeing up the way in which the conditional variances of the common factors affect risk premia can help us to explain returns better. That is, if θbje = θsje = θbji = θsji = θsjm = 0. 1.3. Estimation Method As by construction, the basic model for excess returns that we seek to estimate can be written in compact form as (6) or, in matrix notation, , where (k = e, i, m) is shorthand for τkλkt + fkt = πkt + fkt. Under the assumption of conditional normality, (6) could be estimated for any N countries simultaneously by maximum likelihood (see KSW for details). With three assets per country, and a non‐diagonal time‐varying conditional idiosyncratic covariance matrix, though, this results in a very time‐consuming procedure even for moderately large N. However, estimation could be considerably simplified if we had data on the basis portfolios; see Sentana (1997a). Unfortunately, we do not usually observe directly. Nevertheless, if we construct three diversified (passive) portfolios consisting of currency deposits, bonds and stocks, with excess returns and respectively, a simple method of estimation and testing can still be developed, similar to the one used by Connor and Korajczyk (1988) for the static APT, and more recently by Forni and Reichlin (1998) and Stock and Watson (1998) for macroeconomic time series applications. Given that the idiosyncratic errors are independent in mean from the δs, the approximate zero factor structure assumption on the vjts implies that (7) Therefore, is a full‐rank, time‐invariant transformation of the basis portfolios, whose factor loadings are a linear combination of the corresponding asset factor loadings. Since the scaling of the common factors is free, we can set without loss of generality. Hence, in matrix notation, we can write , with unit lower‐triangular and obviously . If we now add the three portfolios in (7) to the list of 3N assets under consideration, we can factorise the joint likelihood function into the marginal component of , and the conditional components corresponding to all the individual countries, which are given by (8) where the relationship between both sets of parameters is . Let us start with the marginal model for the portfolios. Since the factors are (conditionally) orthogonal, we can decompose the joint log‐likelihood function of (given the past) into the marginal component of plus the conditional of given plus the conditional of given and . This yields (9) In the conditionally homoscedastic case, (9) is a recursive simultaneous equation system, and the parameter estimates are particularly simple to obtain: (a) πe and λe from the OLS regression of on a constant (b) and λi from the OLS regression of on and a constant (c) and λm from the OLS regression of on and a constant. However, there is a generated regressor problem in (c) which affects inferences involving . Nevertheless, consistent standard errors for all the parameters in (9) can be obtained by regarding the above estimation method as GMM based on the just‐identifying moment conditions implicit in (a), (b) and (c) simultaneously. A significant advantage of the GMM framework is that it is easy to see that the estimators for πs and βs remain consistent when fet,fit and fmt follow univariate GARCH processes (and even when the prices of risk are time‐varying), provided that the factor representing portfolios remain contemporaneously uncorrelated, although we must take into account that the residuals will be serially correlated in order to compute consistent standard errors.7 Let us now turn to the conditional models for given . In the conditionally homoscedastic case, ML estimates of can be obtained by using seemingly unrelated regression (SUR) techniques. If the idiosyncratic covariance matrix were block‐diagonal, SUR applied to the 3N assets simultaneously would be equivalent to SUR applied to the three representative assets of each country at a time with as regressors and, in addition, there would be no efficiency loss in estimating the model only for the N countries of interest. Besides, the triangular nature of implies that (iterated) SUR estimates could be obtained simply as follows; see Sentana (1997b): (d) and ωje from the OLS regression of on (e) and ωji from the OLS regression of on and with the residual from (d) as an extra regressor (f) and ωjm from the OLS regression of on and with the residuals from (d) and (e) as extra regressors. Moreover, given the Frisch‐‐Waugh theorem, it is straightforward to prove that the estimates of could be obtained directly if we ran the OLS regressions in (d), (e) and (f) with the orthogonalised portfolios, instead. However, as before, the standard errors associated with and would suffer from a generated regressor bias. Nevertheless, consistent standard errors for all the parameters could be obtained by recasting the above estimation method as GMM based on the just‐identifying moment conditions implicit in (a), (b) and (c) together with (d), (e) and (f) for all countries simultaneously. A significant advantage of the GMM framework is that it is straightforward to conduct inferences involving parameters from different equations that remain valid when some of the assumptions made for estimation, such as block diagonality of the idiosyncratic covariance matrix, do not hold. Furthermore, it is also easy to see that the GMM parameter estimators remain consistent even if ωjet,ωjit and ωjmt follow GARCH processes.8 The different alternative hypotheses discussed above can also be easily tested in this GMM‐regression framework. For instance, to see if idiosyncratic exchange rate variability affects currency risk premia, we can test whether adding the estimated conditional variance of vjet, , as an extra regressor in (6a) improves the explanatory power of the equation. The rationale comes from the fact that (6a) can be rewritten as E. Such a testing procedure also yields as a by‐product an estimate of the coefficient for ωjet under the alternative, ρcje, which is of interest to measure the effects of reducing the variability of idiosyncratic risks. In view of the relatively short number of observations, we choose univariate GARCH (1,1) specifications for the conditional variances of both common and idiosyncratic factors. Such a parametrisation has been found to be a good representation of many financial time series. More efficient conditional variance estimates could be obtained by using a larger information set, or taking into account peso‐type effects. However, note that, even if the GARCH(1,1) model were incorrect, our tests would still be consistent, albeit less powerful, since they are simply rational expectations‐type orthogonality tests, and the GARCH model generates a conditional variance estimate which, conditional on the parameters, is a function of past information. In this respect, it is important to emphasise that our tests will have the correct asymptotic size under the null despite the fact that the estimated conditional variances are generated regressors; see Pagan and Ullah (1988). 2. Empirical Application 2.1. Data We estimate the model described above using monthly data for currency, bond and stock returns (in percentage terms) on ten European countries between October 1977 and October 1997 (ie 241 observations), with the first observation used for initialising the conditional variance recursions. Details of data sources and transformations can be found in the Appendix. The ten countries are Belgium, Denmark, France, Germany, Italy, the Netherlands, Spain, Sweden, Switzerland and the UK. Importantly, note that the returns on both bonds and particularly stocks for each country correspond already to a well‐diversified basket of domestic assets. Since it is crucial that the aggregate portfolios contain the larger non‐European countries as well, we also have data on Australia, Canada, Japan and the US. In this respect, we consider equally weighted world currency, bond and stock returns as our set of portfolios in (7), which are linearly related to the OLS representing portfolios obtained from (6) under the assumption that βajk = 1(a = c, b, s; k = e, i, m) for all the countries for which we have data; see Sentana (2002).9 It is important to mention that Sweden and Switzerland have never belonged to the EMS, although their currencies were unofficially pegged to the DM for part of the sample. The UK only entered the ERM in October 1990, to leave in September 1992 together with Italy, who rejoined the system in November 1996. Finally, Spain, who joined in June 1989, was forced to realign the central parity of its currency four times between September 1992 and March 1995.10 These three countries negotiated a wide margin of ±6% for their bilateral exchange rates, while the other countries, including Italy between 1989 and 1992, participated with bilateral limits of ±2.25% until August 1993, when the bands were widened to ±15%. 2.2. Estimates of the Asset Pricing Model We initially estimated the 195 parameters characterising the model under the null by maximum likelihood as explained in Section 1.3. As a by‐product, we obtained estimates of the conditional variances of common and specific factors. Unfortunately, the computation of consistent standard errors à la Bollerslev and Wooldridge (1992), which avoid the generated regressor problem and are robust to non‐normality, is not an easy task in our case, as it involves the outer product of the gradient and the Hessian matrix of the joint log‐likelihood function for the 33 asset returns. In addition, the outer‐product matrix is singular under some of the alternatives because the number of parameters exceeds the number of observations. For that reason, we only present the GMM estimates discussed in Section 1.3, which turn out to be rather close to the ML ones.11 As for standard errors, we use the Newey and West (1987) formulae with a baseline bandwidth of 6 lags (≃2401/3). Qualitatively similar results are obtained by doubling or halving the lag length. Nevertheless, it is important to emphasise that the finite sample properties of the covariance matrix estimators and associated test procedures in our set‐up are unknown. The results for the subsystem (7) are presented in Table 1. As one would expect from the fact that all returns are US$ denominated, the estimated factor loadings confirm that all three portfolios are positively correlated. Besides, they also confirm that, controlling for exchange rate variations, world bond and stock returns are positively correlated. In this respect, note that since is very close to 1, the interest basis portfolio is hardly distinguishable from an equally weighted average of excess bond returns in local currency. The estimates of the πs, though, are imprecise. The average expected returns on common risks are all positive, but insignificant at conventional levels, with the possible exception of the common interest rate risk if we consider the more relevant one‐sided test.12 Table 1 Prices of Risk and Factor Loadings for Equally‐weighted World Portfolios* World . Common exchange . Common interest . Common stock . Unconditional . portfolio . rate risk . rate risk . market risk . variances . Currencies λe =7.18554 (0.746466) Bonds βbpi =1 λi =1.54193 (0.038047) (0.203489) Stocks βspi =1.33681 βspm =1 λm =11.3880 (0.115252) (0.212905) (3.39880) Unconditional πe =0.00849632 πi =0.219093 πm =0.356913 risk premia (0.198522) (0.126910) (0.243211) World . Common exchange . Common interest . Common stock . Unconditional . portfolio . rate risk . rate risk . market risk . variances . Currencies λe =7.18554 (0.746466) Bonds βbpi =1 λi =1.54193 (0.038047) (0.203489) Stocks βspi =1.33681 βspm =1 λm =11.3880 (0.115252) (0.212905) (3.39880) Unconditional πe =0.00849632 πi =0.219093 πm =0.356913 risk premia (0.198522) (0.126910) (0.243211) Note *Sample period 1977:11‐‐1997:10. GMM estimates of (7): , , , πk= E(πkt), λk= E(λkt)=V(fkt)(k=e,i,m). Newey‐‐West (1987) heteroscedasticity and autocorrelation robust standard errors in parenthesis. Open in new tab Table 1 Prices of Risk and Factor Loadings for Equally‐weighted World Portfolios* World . Common exchange . Common interest . Common stock . Unconditional . portfolio . rate risk . rate risk . market risk . variances . Currencies λe =7.18554 (0.746466) Bonds βbpi =1 λi =1.54193 (0.038047) (0.203489) Stocks βspi =1.33681 βspm =1 λm =11.3880 (0.115252) (0.212905) (3.39880) Unconditional πe =0.00849632 πi =0.219093 πm =0.356913 risk premia (0.198522) (0.126910) (0.243211) World . Common exchange . Common interest . Common stock . Unconditional . portfolio . rate risk . rate risk . market risk . variances . Currencies λe =7.18554 (0.746466) Bonds βbpi =1 λi =1.54193 (0.038047) (0.203489) Stocks βspi =1.33681 βspm =1 λm =11.3880 (0.115252) (0.212905) (3.39880) Unconditional πe =0.00849632 πi =0.219093 πm =0.356913 risk premia (0.198522) (0.126910) (0.243211) Note *Sample period 1977:11‐‐1997:10. GMM estimates of (7): , , , πk= E(πkt), λk= E(λkt)=V(fkt)(k=e,i,m). Newey‐‐West (1987) heteroscedasticity and autocorrelation robust standard errors in parenthesis. Open in new tab The GMM estimates of the factor loadings for each of the three asset classes can be found in Tables 2–4. For clarity of exposition, but without loss of generality, the results for bonds and stocks correspond hereinafter to local currency returns. Since and , the only coefficients affected are those related to the effects of common and specific exchange rate risks. In particular, , , and . Table 2 Factor Loadings for Currency Returns ($)* . Common exchange . Idiosyncratic . Country . rate risk () . variance (ωje) . Belgium 1.26526 0.856507 (0.025079) (0.160474) Denmark 1.22027 0.894656 (0.023217) (0.125631) France 1.20585 0.987137 (0.024049) (0.152557) Germany 1.24651 0.848515 (0.023489) (0.092211) Italy 1.04731 2.17514 (0.047859) (0.606642) Netherlands 1.26122 0.817188 (0.025452) (0.105427) Spain 1.05542 2.40062 (0.049952) (0.602722) Sweden 1.00790 2.87831 (0.056559) (1.06253) Switzerland 1.33764 2.61699 (0.036824) (0.384713) UK 1.02758 4.21787 (0.063866) (0.609914) . Common exchange . Idiosyncratic . Country . rate risk () . variance (ωje) . Belgium 1.26526 0.856507 (0.025079) (0.160474) Denmark 1.22027 0.894656 (0.023217) (0.125631) France 1.20585 0.987137 (0.024049) (0.152557) Germany 1.24651 0.848515 (0.023489) (0.092211) Italy 1.04731 2.17514 (0.047859) (0.606642) Netherlands 1.26122 0.817188 (0.025452) (0.105427) Spain 1.05542 2.40062 (0.049952) (0.602722) Sweden 1.00790 2.87831 (0.056559) (1.06253) Switzerland 1.33764 2.61699 (0.036824) (0.384713) UK 1.02758 4.21787 (0.063866) (0.609914) Note *Sample period 1977:11–1997:10. GMM estimates of (6a): , ωje= E(ωjet)= V(vjet). Newey‐‐West (1987) heteroscedasticity and autocorrelation robust standard errors in parenthesis. Open in new tab Table 2 Factor Loadings for Currency Returns ($)* . Common exchange . Idiosyncratic . Country . rate risk () . variance (ωje) . Belgium 1.26526 0.856507 (0.025079) (0.160474) Denmark 1.22027 0.894656 (0.023217) (0.125631) France 1.20585 0.987137 (0.024049) (0.152557) Germany 1.24651 0.848515 (0.023489) (0.092211) Italy 1.04731 2.17514 (0.047859) (0.606642) Netherlands 1.26122 0.817188 (0.025452) (0.105427) Spain 1.05542 2.40062 (0.049952) (0.602722) Sweden 1.00790 2.87831 (0.056559) (1.06253) Switzerland 1.33764 2.61699 (0.036824) (0.384713) UK 1.02758 4.21787 (0.063866) (0.609914) . Common exchange . Idiosyncratic . Country . rate risk () . variance (ωje) . Belgium 1.26526 0.856507 (0.025079) (0.160474) Denmark 1.22027 0.894656 (0.023217) (0.125631) France 1.20585 0.987137 (0.024049) (0.152557) Germany 1.24651 0.848515 (0.023489) (0.092211) Italy 1.04731 2.17514 (0.047859) (0.606642) Netherlands 1.26122 0.817188 (0.025452) (0.105427) Spain 1.05542 2.40062 (0.049952) (0.602722) Sweden 1.00790 2.87831 (0.056559) (1.06253) Switzerland 1.33764 2.61699 (0.036824) (0.384713) UK 1.02758 4.21787 (0.063866) (0.609914) Note *Sample period 1977:11–1997:10. GMM estimates of (6a): , ωje= E(ωjet)= V(vjet). Newey‐‐West (1987) heteroscedasticity and autocorrelation robust standard errors in parenthesis. Open in new tab Table 3 Factor Loadings for Bond Returns (local currency)* . Common . Common . Specific . Idosyncratic . . exchange rate . interest rate . exchange rate . variance . Country . risk () . risk (βbji) . risk () . (ωji) . Belgium 0.012869 0.746426 −0.010701 1.03904 (0.040965) (0.063506) (0.099932) (0.134862) Denmark −0.031760 1.15055 0.05528 4.34104 (0.052475) (0.098457) (0.135574) (1.02396) France −0.004692 1.02727 0.082677 1.40630 (0.047316) (0.072686) (0.091778) (0.211966) Germany 0.143256 0.929245 0.270939 0.882759 (0.056360) (0.065022) (0.074528) (0.147316) Italy −0.047515 0.962191 0.347770 2.88643 (0.053223) (0.104847) (0.078079) (0.471932) Netherlands 0.069076 0.908067 0.225466 0.863978 (0.042638) (0.059902) (0.077839) (0.153695) Spain −0.086345 0.819648 −0.014769 2.71115 (0.053599) (0.133285) (0.079147) (0.248375) Sweden −0.045976 0.788694 0.060973 2.70035 (0.052172) (0.115046) (0.122362) (0.353565) Switzerland 0.049296 0.590225 0.067478 1.04408 (0.039099) (0.075243) (0.041963) (0.123455) UK 0.056706 1.36035 0.277993 3.94258 (0.056990) (0.109731) (0.092055) (0.564751) . Common . Common . Specific . Idosyncratic . . exchange rate . interest rate . exchange rate . variance . Country . risk () . risk (βbji) . risk () . (ωji) . Belgium 0.012869 0.746426 −0.010701 1.03904 (0.040965) (0.063506) (0.099932) (0.134862) Denmark −0.031760 1.15055 0.05528 4.34104 (0.052475) (0.098457) (0.135574) (1.02396) France −0.004692 1.02727 0.082677 1.40630 (0.047316) (0.072686) (0.091778) (0.211966) Germany 0.143256 0.929245 0.270939 0.882759 (0.056360) (0.065022) (0.074528) (0.147316) Italy −0.047515 0.962191 0.347770 2.88643 (0.053223) (0.104847) (0.078079) (0.471932) Netherlands 0.069076 0.908067 0.225466 0.863978 (0.042638) (0.059902) (0.077839) (0.153695) Spain −0.086345 0.819648 −0.014769 2.71115 (0.053599) (0.133285) (0.079147) (0.248375) Sweden −0.045976 0.788694 0.060973 2.70035 (0.052172) (0.115046) (0.122362) (0.353565) Switzerland 0.049296 0.590225 0.067478 1.04408 (0.039099) (0.075243) (0.041963) (0.123455) UK 0.056706 1.36035 0.277993 3.94258 (0.056990) (0.109731) (0.092055) (0.564751) Note *Sample period 1977:11‐‐1997:10. GMM estimates of (6b): , ωji= E(ωjit)= V(vjit). Newey‐‐West (1987) heteroscedasticity and autocorrelation robust standard errors in parenthesis. Open in new tab Table 3 Factor Loadings for Bond Returns (local currency)* . Common . Common . Specific . Idosyncratic . . exchange rate . interest rate . exchange rate . variance . Country . risk () . risk (βbji) . risk () . (ωji) . Belgium 0.012869 0.746426 −0.010701 1.03904 (0.040965) (0.063506) (0.099932) (0.134862) Denmark −0.031760 1.15055 0.05528 4.34104 (0.052475) (0.098457) (0.135574) (1.02396) France −0.004692 1.02727 0.082677 1.40630 (0.047316) (0.072686) (0.091778) (0.211966) Germany 0.143256 0.929245 0.270939 0.882759 (0.056360) (0.065022) (0.074528) (0.147316) Italy −0.047515 0.962191 0.347770 2.88643 (0.053223) (0.104847) (0.078079) (0.471932) Netherlands 0.069076 0.908067 0.225466 0.863978 (0.042638) (0.059902) (0.077839) (0.153695) Spain −0.086345 0.819648 −0.014769 2.71115 (0.053599) (0.133285) (0.079147) (0.248375) Sweden −0.045976 0.788694 0.060973 2.70035 (0.052172) (0.115046) (0.122362) (0.353565) Switzerland 0.049296 0.590225 0.067478 1.04408 (0.039099) (0.075243) (0.041963) (0.123455) UK 0.056706 1.36035 0.277993 3.94258 (0.056990) (0.109731) (0.092055) (0.564751) . Common . Common . Specific . Idosyncratic . . exchange rate . interest rate . exchange rate . variance . Country . risk () . risk (βbji) . risk () . (ωji) . Belgium 0.012869 0.746426 −0.010701 1.03904 (0.040965) (0.063506) (0.099932) (0.134862) Denmark −0.031760 1.15055 0.05528 4.34104 (0.052475) (0.098457) (0.135574) (1.02396) France −0.004692 1.02727 0.082677 1.40630 (0.047316) (0.072686) (0.091778) (0.211966) Germany 0.143256 0.929245 0.270939 0.882759 (0.056360) (0.065022) (0.074528) (0.147316) Italy −0.047515 0.962191 0.347770 2.88643 (0.053223) (0.104847) (0.078079) (0.471932) Netherlands 0.069076 0.908067 0.225466 0.863978 (0.042638) (0.059902) (0.077839) (0.153695) Spain −0.086345 0.819648 −0.014769 2.71115 (0.053599) (0.133285) (0.079147) (0.248375) Sweden −0.045976 0.788694 0.060973 2.70035 (0.052172) (0.115046) (0.122362) (0.353565) Switzerland 0.049296 0.590225 0.067478 1.04408 (0.039099) (0.075243) (0.041963) (0.123455) UK 0.056706 1.36035 0.277993 3.94258 (0.056990) (0.109731) (0.092055) (0.564751) Note *Sample period 1977:11‐‐1997:10. GMM estimates of (6b): , ωji= E(ωjit)= V(vjit). Newey‐‐West (1987) heteroscedasticity and autocorrelation robust standard errors in parenthesis. Open in new tab Table 4 Factor Loadings for Stock Returns (local currency)* . Common . Common . Common . Specific . Specific . Idiosyncratic . . exchange rate . interest rate . market . exchange rate . interest rate . variance . Country . risk () . risk (βsji ) . risk (βsjm) . risk () . risk (δsji) . (ωjm) . Belgium −0.322000 1.54156 0.869533 0.128528 0.752030 11.0754 (0.1168345) (0.225872) (0.072319) (0.244034) (0.203083) (2.26265) Denmark −0.374027 1.30474 0.642234 −0.247583 0.627680 13.9370 (0.120046) (0.212141) (0.086603) (0.298787) (0.132589) (1.65446) France −0.117952 1.66850 1.19002 0.790233 1.19843 14.4385 (0.183294) (0.300995) (0.069961) (0.378364) (0.268907) (2.19152) Germany −0.239521 1.50536 1.07267 0.774702 0.695732 12.4569 (0.148164) (0.298862) (0.078756) (0.254677) (0.217266) (1.87028) Italy −0.453854 1.37731 1.15685 0.458841 0.575473 30.5493 (0.195227) (0.418395) (0.191447) (0.238861) (0.225265) (4.73570) Netherlands −0.509754 1.56830 0.979499 −0.239149 0.333657 6.80658 (0.143979) (0.227616) (0.064407) (0.176940) (0.165672) (0.932920) Spain −0.264740 1.31521 1.08326 −0.057328 0.724307 19.1805 (0.147007) (0.400857) (0.101797) (0.195961) (0.179790) (2.55937) Sweden −0.493719 1.29952 1.19243 −0.644218 0.570327 21.0370 (0.171656) (0.367956) (0.121713) (0.287745) (0.231128) (2.55007) Switzerland −0.286744 1.56240 0.895178 −0.035626 0.755266 7.66734 (0.146489) (0.260817) (0.079937) (0.137907) (0.158261) (0.856350) UK −0.221163 1.46592 1.01946 −0.122607 0.689536 8.18090 (0.142131) (0.262333) (0.080639) (0.120984) (0.114595) (1.04401) . Common . Common . Common . Specific . Specific . Idiosyncratic . . exchange rate . interest rate . market . exchange rate . interest rate . variance . Country . risk () . risk (βsji ) . risk (βsjm) . risk () . risk (δsji) . (ωjm) . Belgium −0.322000 1.54156 0.869533 0.128528 0.752030 11.0754 (0.1168345) (0.225872) (0.072319) (0.244034) (0.203083) (2.26265) Denmark −0.374027 1.30474 0.642234 −0.247583 0.627680 13.9370 (0.120046) (0.212141) (0.086603) (0.298787) (0.132589) (1.65446) France −0.117952 1.66850 1.19002 0.790233 1.19843 14.4385 (0.183294) (0.300995) (0.069961) (0.378364) (0.268907) (2.19152) Germany −0.239521 1.50536 1.07267 0.774702 0.695732 12.4569 (0.148164) (0.298862) (0.078756) (0.254677) (0.217266) (1.87028) Italy −0.453854 1.37731 1.15685 0.458841 0.575473 30.5493 (0.195227) (0.418395) (0.191447) (0.238861) (0.225265) (4.73570) Netherlands −0.509754 1.56830 0.979499 −0.239149 0.333657 6.80658 (0.143979) (0.227616) (0.064407) (0.176940) (0.165672) (0.932920) Spain −0.264740 1.31521 1.08326 −0.057328 0.724307 19.1805 (0.147007) (0.400857) (0.101797) (0.195961) (0.179790) (2.55937) Sweden −0.493719 1.29952 1.19243 −0.644218 0.570327 21.0370 (0.171656) (0.367956) (0.121713) (0.287745) (0.231128) (2.55007) Switzerland −0.286744 1.56240 0.895178 −0.035626 0.755266 7.66734 (0.146489) (0.260817) (0.079937) (0.137907) (0.158261) (0.856350) UK −0.221163 1.46592 1.01946 −0.122607 0.689536 8.18090 (0.142131) (0.262333) (0.080639) (0.120984) (0.114595) (1.04401) Note *Sample period 1977:11–1997:10. GMM estimates of (6c): , ωjm= E(ωjmt)= V(vjmt). Newey‐‐West (1987) heteroscedasticity and autocorrelation robust standard errors in parenthesis. Open in new tab Table 4 Factor Loadings for Stock Returns (local currency)* . Common . Common . Common . Specific . Specific . Idiosyncratic . . exchange rate . interest rate . market . exchange rate . interest rate . variance . Country . risk () . risk (βsji ) . risk (βsjm) . risk () . risk (δsji) . (ωjm) . Belgium −0.322000 1.54156 0.869533 0.128528 0.752030 11.0754 (0.1168345) (0.225872) (0.072319) (0.244034) (0.203083) (2.26265) Denmark −0.374027 1.30474 0.642234 −0.247583 0.627680 13.9370 (0.120046) (0.212141) (0.086603) (0.298787) (0.132589) (1.65446) France −0.117952 1.66850 1.19002 0.790233 1.19843 14.4385 (0.183294) (0.300995) (0.069961) (0.378364) (0.268907) (2.19152) Germany −0.239521 1.50536 1.07267 0.774702 0.695732 12.4569 (0.148164) (0.298862) (0.078756) (0.254677) (0.217266) (1.87028) Italy −0.453854 1.37731 1.15685 0.458841 0.575473 30.5493 (0.195227) (0.418395) (0.191447) (0.238861) (0.225265) (4.73570) Netherlands −0.509754 1.56830 0.979499 −0.239149 0.333657 6.80658 (0.143979) (0.227616) (0.064407) (0.176940) (0.165672) (0.932920) Spain −0.264740 1.31521 1.08326 −0.057328 0.724307 19.1805 (0.147007) (0.400857) (0.101797) (0.195961) (0.179790) (2.55937) Sweden −0.493719 1.29952 1.19243 −0.644218 0.570327 21.0370 (0.171656) (0.367956) (0.121713) (0.287745) (0.231128) (2.55007) Switzerland −0.286744 1.56240 0.895178 −0.035626 0.755266 7.66734 (0.146489) (0.260817) (0.079937) (0.137907) (0.158261) (0.856350) UK −0.221163 1.46592 1.01946 −0.122607 0.689536 8.18090 (0.142131) (0.262333) (0.080639) (0.120984) (0.114595) (1.04401) . Common . Common . Common . Specific . Specific . Idiosyncratic . . exchange rate . interest rate . market . exchange rate . interest rate . variance . Country . risk () . risk (βsji ) . risk (βsjm) . risk () . risk (δsji) . (ωjm) . Belgium −0.322000 1.54156 0.869533 0.128528 0.752030 11.0754 (0.1168345) (0.225872) (0.072319) (0.244034) (0.203083) (2.26265) Denmark −0.374027 1.30474 0.642234 −0.247583 0.627680 13.9370 (0.120046) (0.212141) (0.086603) (0.298787) (0.132589) (1.65446) France −0.117952 1.66850 1.19002 0.790233 1.19843 14.4385 (0.183294) (0.300995) (0.069961) (0.378364) (0.268907) (2.19152) Germany −0.239521 1.50536 1.07267 0.774702 0.695732 12.4569 (0.148164) (0.298862) (0.078756) (0.254677) (0.217266) (1.87028) Italy −0.453854 1.37731 1.15685 0.458841 0.575473 30.5493 (0.195227) (0.418395) (0.191447) (0.238861) (0.225265) (4.73570) Netherlands −0.509754 1.56830 0.979499 −0.239149 0.333657 6.80658 (0.143979) (0.227616) (0.064407) (0.176940) (0.165672) (0.932920) Spain −0.264740 1.31521 1.08326 −0.057328 0.724307 19.1805 (0.147007) (0.400857) (0.101797) (0.195961) (0.179790) (2.55937) Sweden −0.493719 1.29952 1.19243 −0.644218 0.570327 21.0370 (0.171656) (0.367956) (0.121713) (0.287745) (0.231128) (2.55007) Switzerland −0.286744 1.56240 0.895178 −0.035626 0.755266 7.66734 (0.146489) (0.260817) (0.079937) (0.137907) (0.158261) (0.856350) UK −0.221163 1.46592 1.01946 −0.122607 0.689536 8.18090 (0.142131) (0.262333) (0.080639) (0.120984) (0.114595) (1.04401) Note *Sample period 1977:11–1997:10. GMM estimates of (6c): , ωjm= E(ωjmt)= V(vjmt). Newey‐‐West (1987) heteroscedasticity and autocorrelation robust standard errors in parenthesis. Open in new tab The coefficient on the common exchange rate factor is positive and statistically significant for every currency return (Table 2). Turning to the effect of the common exchange rate on bond returns, note that the coefficients are both positive and negative, but only significantly positive for Germany. This could be because for this country, whose central bank was traditionally concerned about imported inflation, a dollar depreciation leads to lower inflation and, hence, higher bond returns. Table 3, column 2, shows that, in all cases, increases in the common interest rate factor (ie the ‘world’ bond return) are associated with higher European bond returns. Not surprisingly, this effect is always statistically significant. Finally, the idiosyncratic exchange rate factors are significantly positive for Germany, Italy, the Netherlands and the UK. Again, the intuition would be that a local currency appreciation leads to lower inflation and higher bond returns. We next examine the factor loadings for stock returns (Table 4). The coefficients on the common exchange rate are negative in all cases, and significantly so for many countries. Hence, local currency stock returns in Europe suffer when the dollar depreciates. On the other hand, the coefficients on the common interest rate factor suggests that periods of higher world bond returns would tend to be associated with higher European stock market returns. In addition, the coefficients on the common residual market risk factor are all significantly positive. Turning to the effects of the idiosyncratic factors on stock returns, we can see that the coefficients of country‐specific exchange rate factors are positive in some cases and negative in others. In particular, an idiosyncratic local currency appreciation significantly increases stock returns in France and Germany, while decreasing them in Sweden. The fact that, for a given country, the sensitivity of returns to common and idiosyncratic exchange rate movements is different is likely to reflect the structure of its foreign trade. Finally, the results also suggest that controlling for exchange rate variations, the idiosyncratic terms for bonds and stocks are positively and significantly correlated within countries.13 2.3. Direct Effects of ERM on the Cost of Capital Figure 1a displays the average of the conditional standard deviations of the idiosyncratic exchange rate components in ERM and non‐ERM countries. For these purposes, and in line with the consensus view among both academics and practitioners, we only include Belgium, Denmark, France, Germany and the Netherlands in the ERM bloc. Notice that the temporal evolution in both groups is somewhat similar, with a fairly tranquil period between 1983 and 1992, preceded and followed by more volatile ones. Not surprisingly, the effects of the successive EMS crises that began in September 1992 are noticeable in the two series, since the speculative attacks affected all European currencies apart from the Dutch guilder. Nevertheless, the average level and magnitude of the movements are rather different, being substantially smaller for those countries that did not modify the central parity of their currencies against the ECU. Therefore, there is clear evidence that a credible target zone system does reduce exchange rate volatility. Fig. 1. Open in new tabDownload slide Average Conditional Standard Deviation of (a) Idiosyncratic Exchange Rate Factors and (b) Idiosyncratic Interest Rate Factors Fig. 1. Open in new tabDownload slide Average Conditional Standard Deviation of (a) Idiosyncratic Exchange Rate Factors and (b) Idiosyncratic Interest Rate Factors Figure 1b is the analogous picture for the idiosyncratic interest rate components. Except during the first part of the sample, and a brief period between March and May 1990, average interest rate volatility was actually smaller in ERM countries than in non‐ERM ones. At the same time, it also seems that, during turbulent periods in the foreign exchange market, idiosyncratic interest rate volatility increased more for those countries which insisted in maintaining their currencies within the bands. Hence, our results confirm that the reduction in idiosyncratic exchange rate volatility may sometimes be achieved at the expense of increases in local interest rate volatility. Fig. 2. Open in new tabDownload slide Net Effect (a) of Idiosyncratic Exchange Rate Volatility on Currency Returns; and Interest Rate Volatility on Bond Returns; (b) of Idiosyncratic Exchange Rate; and (c) of Idiosyncratic Exchange Rate and Interest Rate Volatility on Stock Returns Fig. 2. Open in new tabDownload slide Net Effect (a) of Idiosyncratic Exchange Rate Volatility on Currency Returns; and Interest Rate Volatility on Bond Returns; (b) of Idiosyncratic Exchange Rate; and (c) of Idiosyncratic Exchange Rate and Interest Rate Volatility on Stock Returns According to our basic asset pricing model, though, such volatility movements should have no impact on expected returns in a world of complete financial market integration. For that reason, we begin by testing whether, contrary to the theory, idiosyncratic exchange rate and interest rate risks are priced. The results for the pricing of idiosyncratic exchange rate volatility in the currency, bond and stock markets of ERM and non‐ERM countries are presented in Table 5 (rows 1 to 3 of panels A and B, respectively). On the currency markets, the joint Wald tests clearly indicate a rejection of the null hypothesis, both in core ERM countries, and especially non‐ERM ones. As for local currency bond and stock returns, the tests would reject in the case of non‐ERM countries, but not in the case of core ERM ones. Table 5 Joint Tests for Pricing of Idiosyncratic Exchange Rate and Interest Rates Risks* Null hypothesis . Risk . Asset . Joint Wald test . p‐value . A: Core ERM countries exchange rate currencies ($) 12.7396 0.02594 exchange rate bonds (l.c.) 7.8618 0.16402 exchange rate stocks (l.c.) 2.2047 0.82016 ρbji =0 ∀ j interest rate bonds 11.7580 0.03826 ρsji =0 ∀ j interest rate stocks 8.3415 0.13839 B: Non‐ERM countries ∀ j exchange rate currencies ($) 47.4100 0.0 ∀ j exchange rate bonds (l.c.) 11.3007 0.04573 ∀ j exchange rate stocks (l.c.) 16.3413 0.00593 ρbji =0 ∀ j interest rate bonds 17.2587 0.00403 ρsji =0 ∀ j interest rate stocks 11.8358 0.03711 Null hypothesis . Risk . Asset . Joint Wald test . p‐value . A: Core ERM countries exchange rate currencies ($) 12.7396 0.02594 exchange rate bonds (l.c.) 7.8618 0.16402 exchange rate stocks (l.c.) 2.2047 0.82016 ρbji =0 ∀ j interest rate bonds 11.7580 0.03826 ρsji =0 ∀ j interest rate stocks 8.3415 0.13839 B: Non‐ERM countries ∀ j exchange rate currencies ($) 47.4100 0.0 ∀ j exchange rate bonds (l.c.) 11.3007 0.04573 ∀ j exchange rate stocks (l.c.) 16.3413 0.00593 ρbji =0 ∀ j interest rate bonds 17.2587 0.00403 ρsji =0 ∀ j interest rate stocks 11.8358 0.03711 Note *Sample period 1977:11–1997:10. Computed using Newey‐‐West (1987) heteroscedasticity and autocorrelation consistent covariance matrix estimator. Open in new tab Table 5 Joint Tests for Pricing of Idiosyncratic Exchange Rate and Interest Rates Risks* Null hypothesis . Risk . Asset . Joint Wald test . p‐value . A: Core ERM countries exchange rate currencies ($) 12.7396 0.02594 exchange rate bonds (l.c.) 7.8618 0.16402 exchange rate stocks (l.c.) 2.2047 0.82016 ρbji =0 ∀ j interest rate bonds 11.7580 0.03826 ρsji =0 ∀ j interest rate stocks 8.3415 0.13839 B: Non‐ERM countries ∀ j exchange rate currencies ($) 47.4100 0.0 ∀ j exchange rate bonds (l.c.) 11.3007 0.04573 ∀ j exchange rate stocks (l.c.) 16.3413 0.00593 ρbji =0 ∀ j interest rate bonds 17.2587 0.00403 ρsji =0 ∀ j interest rate stocks 11.8358 0.03711 Null hypothesis . Risk . Asset . Joint Wald test . p‐value . A: Core ERM countries exchange rate currencies ($) 12.7396 0.02594 exchange rate bonds (l.c.) 7.8618 0.16402 exchange rate stocks (l.c.) 2.2047 0.82016 ρbji =0 ∀ j interest rate bonds 11.7580 0.03826 ρsji =0 ∀ j interest rate stocks 8.3415 0.13839 B: Non‐ERM countries ∀ j exchange rate currencies ($) 47.4100 0.0 ∀ j exchange rate bonds (l.c.) 11.3007 0.04573 ∀ j exchange rate stocks (l.c.) 16.3413 0.00593 ρbji =0 ∀ j interest rate bonds 17.2587 0.00403 ρsji =0 ∀ j interest rate stocks 11.8358 0.03711 Note *Sample period 1977:11–1997:10. Computed using Newey‐‐West (1987) heteroscedasticity and autocorrelation consistent covariance matrix estimator. Open in new tab We also find that idiosyncratic interest rate seems to be priced in bond markets (see row 4 of panels A and B), but in stock markets the evidence against the null is only significant for non‐ERM countries (see row 5). In this respect, it is important to bear in mind that all tests have, ceteris paribus, lower power in the stock return equations because the standard deviation of risks specific to stocks in each country are much higher (compare the idiosyncratic variances in Tables 2‐‐4). Our overall negative results are perhaps not totally surprising in view of the existence of exchange controls (at least for part of the sample) and other barriers to cross‐border investment, as well as informational asymmetries, behavioural biases, liquidity premia etc; (see also Section 2.4). Given that the individual coefficients show great dispersion, being even negative sometimes, and that the effects of exchange rate and interest rate volatility movements on risk premia may partly compensate each other, we have attempted to measure the net effects of idiosyncratic exchange rate and interest rate volatility on each asset by computing the differences in fitted values between alternative and null. Note that, in this way, each country acts as its own control. Figure 2a presents the average net effect on currency returns across ERM and non‐ERM countries, while Figures 2b and 2c display the analogous effects for bond and stock returns respectively. The corresponding sample means and relevant t‐statistics are reported in Table 6. Please note the changes in scale, which reflect the differences in average risk premia across asset classes. Table 6 Net Effect of Idiosyncratic Exchange Rate and Interest Volatility on Returns* Asset . Countries . Average . t‐statistic . p‐value . Currencies ($) Core ERM 0.029019 35.3667 0 Non‐ERM 0.022681 5.0649 0 Difference −0.006338 −1.4525 0.146 Bonds (l.c.) Core ERM −0.022311 −1.5830 0.113 Non‐ERM 0.049634 5.1728 0 Difference 0.071945 5.1430 0 Stocks (l.c.) Core ERM −0.020553 −2.5034 0.012 Non‐ERM 0.065158 2.6136 0.009 Difference 0.085711 3.4089 0.001 Asset . Countries . Average . t‐statistic . p‐value . Currencies ($) Core ERM 0.029019 35.3667 0 Non‐ERM 0.022681 5.0649 0 Difference −0.006338 −1.4525 0.146 Bonds (l.c.) Core ERM −0.022311 −1.5830 0.113 Non‐ERM 0.049634 5.1728 0 Difference 0.071945 5.1430 0 Stocks (l.c.) Core ERM −0.020553 −2.5034 0.012 Non‐ERM 0.065158 2.6136 0.009 Difference 0.085711 3.4089 0.001 Note *Sample period 1977:11–1997:10. Computed using Newey‐‐West (1987) heteroscedasticity and autocorrelation consistent covariance matrix estimator. Open in new tab Table 6 Net Effect of Idiosyncratic Exchange Rate and Interest Volatility on Returns* Asset . Countries . Average . t‐statistic . p‐value . Currencies ($) Core ERM 0.029019 35.3667 0 Non‐ERM 0.022681 5.0649 0 Difference −0.006338 −1.4525 0.146 Bonds (l.c.) Core ERM −0.022311 −1.5830 0.113 Non‐ERM 0.049634 5.1728 0 Difference 0.071945 5.1430 0 Stocks (l.c.) Core ERM −0.020553 −2.5034 0.012 Non‐ERM 0.065158 2.6136 0.009 Difference 0.085711 3.4089 0.001 Asset . Countries . Average . t‐statistic . p‐value . Currencies ($) Core ERM 0.029019 35.3667 0 Non‐ERM 0.022681 5.0649 0 Difference −0.006338 −1.4525 0.146 Bonds (l.c.) Core ERM −0.022311 −1.5830 0.113 Non‐ERM 0.049634 5.1728 0 Difference 0.071945 5.1430 0 Stocks (l.c.) Core ERM −0.020553 −2.5034 0.012 Non‐ERM 0.065158 2.6136 0.009 Difference 0.085711 3.4089 0.001 Note *Sample period 1977:11–1997:10. Computed using Newey‐‐West (1987) heteroscedasticity and autocorrelation consistent covariance matrix estimator. Open in new tab If we ignore the negative spikes, which correspond to devaluations in the Swedish krona, the evidence on currency returns suggests that, for both groups of countries, idiosyncratic exchange rate variability significantly increases short interest rate differentials for a given expected depreciation. This could explain the convergence of short rates in Europe during periods of low intra‐European exchange rate volatility. Therefore, a system that reduced both the expected level and the volatility of idiosyncratic exchange rate movements would seem to reduce the riskless component of the cost of capital. The evidence for bond and stock markets is less clear cut, with a small negative effect in the ERM bloc on average over time, and a significantly larger, positive effect in the other group. Nevertheless, a comparison of Figures 2b and c with Figure 1a suggests that periods in which local exchange rate volatility has risen have tended to be associated with increases in the required rate of return on bonds and equities. If with all the usual caveats in place, we were to extrapolate these results to a currency system such as EMU, in which country‐specific exchange rate uncertainty has almost vanished, and the interest rate policy of the member countries is not only common, but more importantly, released from the obligation of maintaining intra‐European exchange rate stability, our tentative conclusion would be that, ceteris paribus, the cost of capital in euroland is likely to be lower than it would have been otherwise, although the gains would probably be fairly small. 2.4. The Gains from Globalisation The international asset pricing model described in Section 1.1 implicitly assumes that European financial markets are integrated. Specifically, we have assumed that idiosyncratic risk is not priced, and that the price of risk associated with each of the underlying common factors is the same for all countries. Therefore, one way of testing for market integration is to examine if idiosyncratic factors are priced. We have already done this in the previous section for exchange rate and interest rate risks. The results of the tests for the pricing of country‐specific residual market risk in the equity markets show that the coefficients are not jointly significantly different from zero (Table 7, row 1), although again this is probably due to lack of power. Table 7 Additional Joint Tests for Market Integration* Null hypothesis . Risk . Asset . Joint Wald test . p‐value . ρsjm =0 ∀ j Residual market Stocks 13.1805 0.21376 Exchange rate Currencies ($) 16.1045 0.09668 Exchange rate Bonds (l.c.) 31.3994 0.00050 Exchange rate Stocks (l.c.) 103.3475 0.0 θbji =0 ∀ j Interest rate Bonds 11.2748 0.33651 θsji =0 ∀ j Interest rate Stocks 26.1813 0.00350 θsjm =0 ∀ j Residual market Stocks 142.2266 0.0 Null hypothesis . Risk . Asset . Joint Wald test . p‐value . ρsjm =0 ∀ j Residual market Stocks 13.1805 0.21376 Exchange rate Currencies ($) 16.1045 0.09668 Exchange rate Bonds (l.c.) 31.3994 0.00050 Exchange rate Stocks (l.c.) 103.3475 0.0 θbji =0 ∀ j Interest rate Bonds 11.2748 0.33651 θsji =0 ∀ j Interest rate Stocks 26.1813 0.00350 θsjm =0 ∀ j Residual market Stocks 142.2266 0.0 Note *Sample period 1977:11–1997:10. Computed using Newey‐‐West (1987) heteroscedasticity and autocorrelation consistent covariance matrix estimator. Open in new tab Table 7 Additional Joint Tests for Market Integration* Null hypothesis . Risk . Asset . Joint Wald test . p‐value . ρsjm =0 ∀ j Residual market Stocks 13.1805 0.21376 Exchange rate Currencies ($) 16.1045 0.09668 Exchange rate Bonds (l.c.) 31.3994 0.00050 Exchange rate Stocks (l.c.) 103.3475 0.0 θbji =0 ∀ j Interest rate Bonds 11.2748 0.33651 θsji =0 ∀ j Interest rate Stocks 26.1813 0.00350 θsjm =0 ∀ j Residual market Stocks 142.2266 0.0 Null hypothesis . Risk . Asset . Joint Wald test . p‐value . ρsjm =0 ∀ j Residual market Stocks 13.1805 0.21376 Exchange rate Currencies ($) 16.1045 0.09668 Exchange rate Bonds (l.c.) 31.3994 0.00050 Exchange rate Stocks (l.c.) 103.3475 0.0 θbji =0 ∀ j Interest rate Bonds 11.2748 0.33651 θsji =0 ∀ j Interest rate Stocks 26.1813 0.00350 θsjm =0 ∀ j Residual market Stocks 142.2266 0.0 Note *Sample period 1977:11–1997:10. Computed using Newey‐‐West (1987) heteroscedasticity and autocorrelation consistent covariance matrix estimator. Open in new tab The other way to assess if European markets are integrated is to test whether the prices of risk associated with the common factors are the same across countries in each of the three asset classes. The null hypothesis of common exchange rate risk evaluation is not rejected in the currency markets (Table 7, row 2), but it is strongly rejected in the bond and stock markets (rows 3 and 4). As for the common interest rate factor, the assumption of common valuation is not rejected in European bond markets (see row 5), but it is rejected in stock markets (row 6). Finally, we also find significant differential pricing for common residual market risks (see row 7). One attractive reinterpretation of our lack of integration results is to say that international investors are incompletely diversified, which is consistent with the well known fact that most investors show a home bias, in the sense that they hold only a relatively small proportion of their assets outside their country; see, for example, Kang and Stulz (1997) for a discussion and references. In this respect, an important indirect effect of EMU would be the elimination of some of the elements that limit cross‐border investments in Europe, such as the restrictions on foreign asset holdings by pension funds, insurance companies and other financial institutions, or the behavioural biases traditionally shown by individual investors in favour of assets denominated in their own domestic currency. In fact, there is ample anecdotal evidence that such a process is already taking place at a very rapid rate. To gauge the potential gains from increased market integration, we are going to follow Stulz (1999) in comparing stock market risk premia under full integration with the risk premia that would prevail in the context of fully segmented markets. In this sense, note that, although the distinction between common and specific risk becomes irrelevant in the latter context, it is still possible to write the unanticipated components of stock returns in terms of three orthogonal sources of risk: (10) Since the stock portfolio for each country corresponds to a diversified basket of domestic stocks, an argument similar to the one presented in Section 1.1 would then result in a domestic version of the APT (11) where ϕje, ϕji and ϕjm are the prices of risk in country j. Note that (11) is a special case of our general alternative hypothesis (5), with the restrictions ρsjk=ϕsjk=τk and θsjk=ϕsjkβjsk(βjsk−1) for k=e,i,m. Assuming as in Stulz (1999) that financial market integration does not affect the prices of residual market risk, so that ϕjm=τm, we can assess whether there would be gains (on average over time) to each country from stock market integration by comparing these quantities:14 Table 8 reports the difference between the right‐ and left‐hand sides of the above expression for the European countries in our database. Apart from the fact that all the differences are highly significantly positive, the other striking result is that there is substantial variation across countries. In particular, the largest average gains correspond to countries such Italy, Sweden or Spain, which tend to have both large βsjms , and especially, very large idiosyncratic variances. As a result, those countries are the ones that a priori would benefit most from an increase in stock market integration. In contrast, the benefits for countries such as the Netherlands, Switzerland or the UK, whose stock markets have significantly smaller residual market risk, probably because they already have closer links with world markets, would be smaller. If we multiply those differences by 0.03, which is roughly our implicitly estimate of τm (see footnote 10), our results suggest that the potential gains from stock market integration could be rather large. Nevertheless, it is important to emphasise that those gains should only be taken as indicative, in view of the large standard error associated with the price of risk coefficient, and the fact that we are comparing a situation of full segmentation with another of complete integration. In practice, of course, markets are neither fully segmented, nor fully integrated and, moreover, the transition from one state to the other is typically a gradual process, whose effects are partly anticipated by investors. In this respect, Hardouvelis et al. (2000) find that the degree of integration of European stock markets increased substantially after 1995, when forward interest differentials vis‐ à‐vis Germany began to narrow in anticipation of EMU membership. Table 8 Average Gains from Increased Stock Market Integration (local currency) * Country . . βsjmλm . Difference . Belgium 19.6857 9.90221 9.78350 (4.43808) (3.42981) (2.30608) Denmark 18.6341 7.31373 11.3204 (2.06554) (1.74589) (2.02143) France 30.5656 13.5519 17.0136 (5.52645) (3.74877) (2.84319) Germany 25.5600 12.2155 13.3445 (5.68106) (4.08642) (2.40860) Italy 45.7898 13.1742 32.6156 (6.33277) (2.51710) (5.46835) Netherlands 17.7324 11.1545 6.57790 (3.73677) (3.32912) (1.21043) Spain 32.5438 12.3362 20.2077 (5.89328) (3.97726) (3.08184) Sweden 37.2293 13.5793 23.6500 (5.21221) (3.48773) (3.14164) Switzerland 16.7930 10.1943 6.59876 (4.33630) (3.78708) (1.06078) UK 20.0163 11.6095 8.40678 (5.49114) (4.02881) (1.70797) Country . . βsjmλm . Difference . Belgium 19.6857 9.90221 9.78350 (4.43808) (3.42981) (2.30608) Denmark 18.6341 7.31373 11.3204 (2.06554) (1.74589) (2.02143) France 30.5656 13.5519 17.0136 (5.52645) (3.74877) (2.84319) Germany 25.5600 12.2155 13.3445 (5.68106) (4.08642) (2.40860) Italy 45.7898 13.1742 32.6156 (6.33277) (2.51710) (5.46835) Netherlands 17.7324 11.1545 6.57790 (3.73677) (3.32912) (1.21043) Spain 32.5438 12.3362 20.2077 (5.89328) (3.97726) (3.08184) Sweden 37.2293 13.5793 23.6500 (5.21221) (3.48773) (3.14164) Switzerland 16.7930 10.1943 6.59876 (4.33630) (3.78708) (1.06078) UK 20.0163 11.6095 8.40678 (5.49114) (4.02881) (1.70797) Note * Sample period 1977:11–1997:10. Newey‐‐West (1987) heteroscedasticity and autocorrelation robust standard errors in parenthesis. Open in new tab Table 8 Average Gains from Increased Stock Market Integration (local currency) * Country . . βsjmλm . Difference . Belgium 19.6857 9.90221 9.78350 (4.43808) (3.42981) (2.30608) Denmark 18.6341 7.31373 11.3204 (2.06554) (1.74589) (2.02143) France 30.5656 13.5519 17.0136 (5.52645) (3.74877) (2.84319) Germany 25.5600 12.2155 13.3445 (5.68106) (4.08642) (2.40860) Italy 45.7898 13.1742 32.6156 (6.33277) (2.51710) (5.46835) Netherlands 17.7324 11.1545 6.57790 (3.73677) (3.32912) (1.21043) Spain 32.5438 12.3362 20.2077 (5.89328) (3.97726) (3.08184) Sweden 37.2293 13.5793 23.6500 (5.21221) (3.48773) (3.14164) Switzerland 16.7930 10.1943 6.59876 (4.33630) (3.78708) (1.06078) UK 20.0163 11.6095 8.40678 (5.49114) (4.02881) (1.70797) Country . . βsjmλm . Difference . Belgium 19.6857 9.90221 9.78350 (4.43808) (3.42981) (2.30608) Denmark 18.6341 7.31373 11.3204 (2.06554) (1.74589) (2.02143) France 30.5656 13.5519 17.0136 (5.52645) (3.74877) (2.84319) Germany 25.5600 12.2155 13.3445 (5.68106) (4.08642) (2.40860) Italy 45.7898 13.1742 32.6156 (6.33277) (2.51710) (5.46835) Netherlands 17.7324 11.1545 6.57790 (3.73677) (3.32912) (1.21043) Spain 32.5438 12.3362 20.2077 (5.89328) (3.97726) (3.08184) Sweden 37.2293 13.5793 23.6500 (5.21221) (3.48773) (3.14164) Switzerland 16.7930 10.1943 6.59876 (4.33630) (3.78708) (1.06078) UK 20.0163 11.6095 8.40678 (5.49114) (4.02881) (1.70797) Note * Sample period 1977:11–1997:10. Newey‐‐West (1987) heteroscedasticity and autocorrelation robust standard errors in parenthesis. Open in new tab 3. Conclusions In this paper, we use monthly data on currency, bond and stock returns for ten European countries over the period 1977–97 to estimate a dynamic multi‐factor APT model with time‐varying volatility in both common and idiosyncratic factors. From the methodological point of view, our main contribution is to exploit the cross‐sectional dimension of the model to construct diversified world portfolios which, when added to the original list of assets, can be used to develop simple, but nevertheless consistent GMM‐based methods of estimation and testing. We find that, controlling for exchange rate variations, world bond and stock returns are positively correlated. We also find that the average expected returns on the common risks are all positive, but not very precisely estimated. Moreover, our results show that increases in the common exchange rate, interest rate and residual market factors are associated with higher European currency, bond and stock returns respectively. They also suggest that a local currency appreciation generally leads to higher (local currency) bond returns, while (local currency) stock returns in Europe suffer when the dollar depreciates. In addition, we find that controlling for exchange rate variations, the idiosyncratic terms for bonds and stocks are positively and significantly correlated within countries. Furthermore, our findings indicate that a target zone system such as the EMS does reduce exchange rate volatility as long as it remains credible, and that average interest rate volatility has actually been smaller in ERM countries than in non‐ERM ones. At the same time, our results confirm that reductions in idiosyncratic exchange rate volatility may sometimes be achieved at the expense of increases in local interest rate volatility. Importantly, our evidence also suggests that a system that reduces both the expected level and the volatility of idiosyncratic exchange rate movements is likely to reduce the riskless component of the cost of capital. The evidence for bond and stock markets is less clear cut, although periods in which local exchange rate volatility has risen have tended to be associated with increases in the required rate of return on bonds and equities. Nevertheless, the effects that we uncover are small. We also find overall negative results on market integration. For that reason, and given that an important indirect effect of EMU would be an increase in the degree of integration of European financial markets, we compare stock market risk premia under full integration with the risk premia that would prevail in the context of fully segmented markets. Our results suggest that such an upper bound on the potential gains from stock market integration could be rather large. Finally, it is worth bearing in mind that there are other important channels through which the removal of financial market segmentation by means of regional arrangements such as the EMS could lead to a decrease in funding costs. Given the separation of ownership and control that exists in the quoted corporate sector, the most important mechanism will be the reduction in the agency costs faced by firms raising outside capital, which are not directly related to the required rate of return used in present discounted value calculations; see Stulz (1999). At the same time, the creation of a wider, deeper and more liquid capital market is likely to intensify the pressure on European firms to pay more attention to shareholders, and to seek to raise the value of their stocks. In this respect, cross‐border mergers and acquisitions show a clear upward trend. In addition, increased competition by banks and other financial institutions may also result in a reduction in the cost of capital for smaller unquoted companies, which account for a significant proportion of European investment. Therefore, there is little doubt that the measuring of the effects of ‘globalisation’ on the cost of capital would continue to be an area of active research. Footnotes 1 " See Ando and Auerbach (1988) for a discussion of the difficulties induced by differential taxation in cross‐country comparisons of the debt component of the cost of capital. 2 " Strictly speaking, excess returns should be defined in terms of arithmetic returns, rather than geometric returns. Nevertheless, the approximation error is usually irrelevant in empirical work; see McCulloch (1975) . 3 " See Sentana and Fiorentini (2001) for the implications of this assumption on the identification of the factors and their loadings. 4 " Their definition is asymptotic, and amounts in our case to the largest eigenvalue of the (conditional) idiosyncratic covariance matrix remaining uniformly bounded as the number of countries , N , goes to infinity. 5 " If the stochastic discount factor ξt contains some idiosyncratic risk, the APT expression (4) becomes approximate, as in Ross’ (1976) original formulation, and it is only possible to prove that the pricing errors would be negligible on average cross‐sectionally, but not necessarily so for each asset. Connor (1984) provides conditions that guarantee exact asset pricing in a competitive equilibrium set‐up. 6 " Ross’ (1976) results were implicitly derived for a closed economy. However, if exchange rates belong to the set of factors, then the APT can be readily generalised to an international setting; see Solnik (1983) and Ikeda (1991) . 7 " Maximum likelihood estimates of the trivariate 3‐factor GARCH in mean model for in (9) can be easily obtained by replacing OLS with univariate GARCH‐M regressions in (a), (b) and (c) above, but the standard errors usually computed would be inconsistent due to the generated regressor problem. 8 " Maximum likelihood estimates of the factor loadings for each country model in (8) can be obtained by replacing OLS in (d), (e) and (f) with univariate GARCH regressions, but again the standard errors usually computed would be inconsistent because of the generated regressor problem. 9 " It is in principle possible, albeit non‐trivial, to estimate the optimal weightings of the basis portfolios together with all the other parameters; (see for example KSW). In any case, note that the consistency of the estimation methods described in Section 1.3 would not be affected if the international portfolios that we construct were not fully diversified, although it would be necessary to develop an asymptotic theory in which both N and T increase at possibly the same rate in order to robustify our inference. Both issues are left for further research. 10 " A complete list of ERM realignments between March 1979 and December 1994 can be found in Ayuso and Pérez‐Jurado (1997) . 11 " A useful way to measure the ‘distance’ between ML and GMM parameter estimates is to look at the differences between the implied basis portfolios for common and specific factors. In this respect, we find that both sets of estimates yield very similar results, with correlations ranging from 0.979 to 1, with an average value of 0.997. 12 " If we compute ‘average’ price of risk coefficients for k = e , i , m as , we obtain 0.001177, 0.140906 and 0.030949, which although strictly speaking inconsistent, are rather similar to the corresponding ML estimates (0.001627, 0.121238 and 0.028841). 13 " As discussed in Section 1.3, our estimation procedure is based on the maintained assumption that the covariance matrix of the idiosyncratic terms satisfies the approximate zero factor structure of Chamberlain and Rothschild (1983). We have informally assessed whether such an assumption is correct on the basis of the correlation matrix of the estimated idiosyncratic terms for all the countries in our dataset. In this respect, we find that the average of its off‐diagonal elements (squared) is 0.017518. Similarly, we find that its largest eigenvalue only explains 10.5% of its trace, as opposed to 13% when we only consider the European countries. 14 " Given the definition of beta as the ratio of covariance to variance, and the fact that for innovations the unconditional variance coincides with the average of the conditional one, we can equivalently write this expression in terms of covariances, as in Stulz (1999) , yielding * " This is a thoroughly revised version of Sentana et al. (1995) , which extended the analysis of stock returns in Sentana et al. (1992) to currencies and bonds. I am very grateful to Mushtaq Shah and Sushil Wadhwani for their contributions to the earlier versions of this paper. I would also like to thank Manuel Arellano, Juan Ayuso, Philippe Bacchetta, José M. Campa, Greg Connor, Charles Goodhart, Mike Wickens and participants at seminars at AUEB (Athens), Bank of England, Bank of Spain, the Board of Governors of the Federal Reserve, Fundación BBV (Bilbao), IAE (Barcelona), LSE, UIMP (Santander), and the 1995 Royal Economic Society Conference for their comments. The editor and three anonymous referees have also helped me greatly to improve the paper. Of course, the usual caveat applies. Thanks are also due to Juan Ayuso, Martin Brookes, Aelred Conelly, Simon Cook, José A. Cuenca and Andrew Haldane for their help in collecting the data and to Isabel Figuerola‐Ferretti and Angel León for research assistance. This work was initiated at the LSE Financial Markets Group, whose financial support for the first version as part of the ESRC project ‘The Efficiency and Regulation of Financial Markets’ is gratefully acknowledged. References Ando , A. and Auerbach , A. J. ( 1988 ). ‘The cost of capital in the United States and Japan: a comparison’ , Journal of the Japanese and International Economy , vol. 2 , pp. 134 – 58 . Google Scholar Crossref Search ADS WorldCat Ayuso , J. and Pérez‐Jurado , M. ( 1997 ). ‘Devaluations and depreciation expectations in the EMS’ , Applied Economics , vol. 29 , pp. 471 – 84 . Google Scholar Crossref Search ADS WorldCat Batchelor , R. ( 1985 ). Evidence to the House of Lords Select Committee. Bekaert , G. and Hodrick , R. J. ( 1992 ). ‘Characterizing predictable components in excess returns on equity and foreign exchange markets’ , Journal of Finance , vol. 47 , pp. 467 – 509 . Google Scholar Crossref Search ADS WorldCat Bollerslev , T. and Wooldridge , J. M. ( 1992 ). ‘Quasi‐maximum likelihood estimation and inference in dynamic models with time‐varying variances’ , Econometric Reviews , vol. 11 , pp. 143 – 72 . Google Scholar Crossref Search ADS WorldCat Cochrane , J. H. ( 2001 ). Asset Pricing . Princeton: Princeton University Press . Google Scholar Google Preview OpenURL Placeholder Text WorldCat COPAC Connor , G. ( 1984 ). ‘A unified beta pricing theory’ , Journal of Economic Theory , vol. 34 , pp. 13 – 31 . Google Scholar Crossref Search ADS WorldCat Connor , G. and Korajczyk , R. A. ( 1988 ). ‘Risk and return in an equilibrium APT: an application of a new test methodology’ , Journal of Financial Economics , vol. 21 , pp. 255 – 89 . Google Scholar Crossref Search ADS WorldCat European Commission ( 1990 ). ‘One market, one money’ , European Economy special issue , October. Google Scholar Google Preview OpenURL Placeholder Text WorldCat COPAC Forni , M. and Reichlin , L. ( 1998 ). ‘Let's get real: a factor analytical approach to disaggregated business cycle dynamics’ , Review of Economic Studies vol. 65 , pp. 453 – 73 . Google Scholar Crossref Search ADS WorldCat Hardouvelis , G. , Malliaropulos , D. and Priestley , R. ( 2000 ). ‘EMU and European stock market integration’ , mimeo, Norwegian School of Management. Ikeda , S. ( 1991 ). ‘Arbitrage asset pricing under exchange risk’ , Journal of Finance , vol. 46 , pp. 447 – 55 . Google Scholar Crossref Search ADS WorldCat Jarrow , R. , Maximovich , M. and Ziemba , W. ( 1995 ). Handbook of Modern Finance , Amsterdam: North Holland. Google Scholar Google Preview OpenURL Placeholder Text WorldCat COPAC Kang , J. ‐K. and Stulz , R. ( 1997 ). ‘Why is there a home bias? An analysis of foreign portfolio equity ownership in Japan’ , Journal of Financial Economics , vol. 46 , pp. 3 – 28 . Google Scholar Crossref Search ADS WorldCat King , M. , Sentana , E. and Wadhwani , S. ( 1994 ). ‘Volatility and links between national stock markets’ , Econometrica , vol. 62 , pp. 901 – 33 . Google Scholar Crossref Search ADS WorldCat McCulloch , J. H. ( 1975 ). ‘Operational aspects of the Siegel paradox’ , Quarterly Journal of Economics , vol. 89 , pp. 170 – 2 . Google Scholar Crossref Search ADS WorldCat Newey , W. and West , K. D. ( 1987 ). ‘A simple positive semi‐definite heteroskedasticity and autocorrelation consistent covariance matrix’ , Econometrica , vol. 55 , pp. 703 – 6 . Google Scholar Crossref Search ADS WorldCat Pagan , A. and Ullah , A. ( 1988 ). ‘The econometric analysis of models with risk terms’ , Journal of Applied Econometrics , vol. 3 , pp. 87 – 105 . Google Scholar Crossref Search ADS WorldCat Ross , S. A. ( 1976 ). ‘The arbitrage theory of capital asset pricing’ , Journal of Economic Theory , vol. 13 , pp. 641 – 60 . Google Scholar Crossref Search ADS WorldCat Sentana , E. ( 1997a ). ‘Risk and return in the Spanish stock market: some evidence from individual assets’ , Investigaciones Económicas , vol. 21 , pp. 297 – 359 . OpenURL Placeholder Text WorldCat Sentana , E. ( 1997b ). ‘Estimation of a triangular seemingly unrelated regression system by OLS’ , Econometric Theory , vol. 13 , Problem 97.2.2. OpenURL Placeholder Text WorldCat Sentana , E. ( 2002 ). ‘Factor representing portfolios in large asset markets’ , forthcoming in the Journal of Econometrics . OpenURL Placeholder Text WorldCat Sentana , E. and Fiorentini , F. ( 2001 ). ‘Identification, estimation and testing of conditionally heteroskedastic factor models’ , Journal of Econometrics , vol. 102 , pp. 143 – 64 . Google Scholar Crossref Search ADS WorldCat Sentana , E. , Shah , M. and Wadhwani , S. ( 1992 ). ‘Has the EMS reduced the cost of capital?’ , LSE FMG Discussion Paper 134 . Sentana , E. , Shah , M. and Wadhwani , S. ( 1995 ). ‘Has the EMS reduced the cost of capital?’ , CEMFI Working Paper 9514 . Shiller , R. J. ( 1979 ). ‘The volatility of long‐term interest rates and expectations models of the term structure’ , Journal of Political Economy , vol. 87 , pp. 1190 – 219 . Google Scholar Crossref Search ADS WorldCat Solnik , B. ( 1983 ). ‘International arbitrage pricing theory’ , Journal of Finance , vol. 18 , pp. 449 – 57 . Google Scholar Crossref Search ADS WorldCat Stock , J. H. and Watson , M. W. ( 1998 ). ‘Diffusion indexes’ , NBER Working Paper 6702 . Stulz , R. ( 1995 ). ‘International asset pricing: an integrative survey’ , in Jarrow et al . (1995), pp. 201 – 23 . Stulz , R. ( 1999 ). ‘Globalization of equity markets and the cost of capital’ , NBER Working Paper 7021 . Appendix Data Appendix Details of the data series used are as follows (where appropriate the name of the series is followed by its Datastream code). Stock prices and dividend yields: *Morgan Stanley Capital International Perspectives Exchange rates: *US$‐Aus$, end of period (AUOCEXCH) *US$‐BelFr, end of period (BGOCEXCH) *US$‐CAN$, end of period (CNOCEXCH) *US$‐DENKr, end of period (DKOCEXCH) *US$‐FraFr, end of period (FROCEXCH) *US$‐DM, end of period (BDOCEXCH) *US$‐ItaLit, end of period (ITOCEXCH) *US$‐Yen, end of period (JPOCEXCH) *US$‐NetFl, end of period (NLOCEXCH) *US$‐SpPta, end of period (ESOCEXCH) *US$‐SweKr, end of period (SDOCEXCH) *US$‐SwiFr, end of period (SWOCEXCH) *US$‐Stg, end of period (UKOCEXCH) Short interest rates: *Belgium, Canada, Denmark, France, Germany, Italy, Japan, Netherlands, Sweden, Switzerland, UK, US: 1‐month Eurocurrency rate from the Bank of England. *Australia: Yield on 3‐month Treasury Bill (AUTRSBL%) *Spain: pre 1983:1: 3‐month interbank rate from the Bank of Spain; post 1983:1: 1‐month Eurocurrency rate. Bond returns: *Canada, France, Germany, Japan, Netherlands, Switzerland, UK, US: pre 1978:1: Shiller (1979) approximation on 10‐year bond yields from Goldman Sachs; post 1978:2: Salomon Brothers bond return indices. *Australia: pre‐1984:10: Shiller (1979) approximation on 10‐year bond yield; post 1984:11: Salomon Brothers bond return indices *Italy: pre 1990:12: Shiller (1979) approximation on 10 year bond yields from Goldman Sachs; post 1991:1: Salomon Brothers bond return indices. *Belgium, Denmark, Sweden: pre 1985:3: Shiller (1979) approximation on 10 year bond yields from Goldman Sachs; 1985:4‐‐1990:12: Datastream bond return indices (Belgium (ABGGYG4(RI)), Denmark (ADKGYG4(RI)), Sweden (ASDGYG4(RI)); post 1991:1: Salomon Brothers bond return indices. *Spain: pre 1982:1: Shiller (1979) approximation on bond yields from Bank of Spain; 1982:2‐1985:3: Shiller (1979) approximation on bond yields from Goldman Sachs; 1985:3:1990:12: Datastream bond return indices (AESGYG4(RI)); post 1991:1: Salomon Brothers bond return indices. © Royal Economic Society 2002

Journal

The Economic Journal – Oxford University Press

Published: Oct 1, 2002

Get 20M+ Full-Text Papers For Less Than $1.50/day. Start a 7-Day Trial for You or Your Team.

Learn More →

Did the EMS Reduce the Cost of Capital?

Did the EMS Reduce the Cost of Capital?

Get 20M+ Full-Text Papers For Less Than $1.50/day. Start a 7-Day Trial for You or Your Team.

Learn More →

Did the EMS Reduce the Cost of Capital?

Did the EMS Reduce the Cost of Capital?

References (29)

Abstract

Journal

Recommended Articles

There are no references for this article.

Our policy towards the use of cookies