Asset prices and exchange risk: Empirical evidence from Canada

Research in International Business and Finance 28 (2013) 35– 44

Contents lists available at SciVerse ScienceDirect

Research in International Businessand Finance

journal homepage: www.elsevier.com/locate/r ibaf

Asset prices and exchange risk: Empirical evidence fromCanada

Lucie Samson ∗

Université Laval, 1025 av. des Sciences-Humaines, Pavillon J.-A. deSève, Québec, Québec, Canada C1V 0A6

a r t i c l e i n f o

Article history:Received 29 June 2010Received in revised form 27 September2012Accepted 28 September 2012

Available online 8 October 2012

Keywords:Asset pricingRisk premiumExchange riskEconomic factors

JEL classification:G10G11G12G15

a b s t r a c t

Asset prices have been found to respond to unpredicted changes inmacroeconomic variables in a number of studies. This paper focuseson the relationship between economic factors and the stock mar-ket for a small open economy, namely Canada. Exchange risk isobserved to have a significant impact on firm value in that countrybetween 1971 and 2004. Inflation risk also played a non negligiblerole during the seventies and eighties. The role played by marketrisk is harder to ascertain.

© 2012 Elsevier B.V. All rights reserved.

1. Introduction

Preference-free valuation models explain the behavior of asset returns through their link with rele-vant risk factors. These factors are those that cannot be diversified away. In some cases, the underlyingrisks are treated as unobserved and the time-varying expected returns are linked to the behavior ofone or more latent variables. In other cases, observable macroeconomic economic variables are usedas factors to capture some of the systematic risks present in the economy. Examples of these variablesare the inflation rate, a market return index, a production index, a measure of aggregate consumption,

∗ Fax: +1 418 656 2707.E-mail address: [email protected]

0275-5319/$ – see front matter © 2012 Elsevier B.V. All rights reserved.http://dx.doi.org/10.1016/j.ribaf.2012.09.006

dx.doi.org/10.1016/j.ribaf.2012.09.006

http://www.sciencedirect.com/science/journal/02755319

http://www.elsevier.com/locate/ribaf

mailto:[email protected]

dx.doi.org/10.1016/j.ribaf.2012.09.006

36 L. Samson / Research in International Business and Finance 28 (2013) 35– 44

the real rate of interest, and a measure of spread between a long and a short rate or between the returnon risky and safe assets.

Two approaches can be used to model the impact of economic variables on asset prices. On onehand, the factor loadings can be treated as constant over time and a simultaneously equation model canbe estimated. When the risk factors are made explicit, the result is a highly constrained model. Whenthey are left unspecified, the resulting latent variable model imposes fewer constraints, is not as easilyrejected, but the source of risk cannot be identified. In the former case, the part of a return that is notexplained by the factor realizations times their factor loading is simply the idiosyncratic risk associatedwith any given asset. Examples of this type of analysis are found in Ferson and Harvey (1991), Eltonet al. (1995), and Harvey et al. (2002) among others. The obvious advantage of this procedure is thatall equations are estimated simultaneously, so there is no errors-in-variables problem. The hypothesisof constant factor loadings over long periods of time can be too strong in some cases however.

Fama and Macbeth (1973), Chen et al. (1986), Fama and French (1992), Shanken and Weinsten(2006), Virk (2012) and others use a cross-sectional procedure that presents the opposite problem.The loadings, the betas, are usually estimated from a series of rolling regressions, allowing them to varyover time, or from the entire sample period, implying that they are kept constant over time. The choicebetween the two procedures is often dictated by the length of the overall sample period, the rollingwindow method necessitating more observations. The betas are subsequently used as regressors incross-sectional regressions to determine the size of various risk premia. Since the same data is used togenerate the betas and estimate the risk premia, the commonly known errors-in-variables problemis then present and the standard errors associated with the premia are usually underestimated. Theresults presented in this paper combine both approaches, with specified risk factors, making it aninteresting check on the robustness of our empirical findings.

The Capital Asset Pricing Model (CAPM) identifies the correlation between the return on any givenasset j and the return on the market portfolio as the relevant factor in the determination of the asso-ciated risk premium. It has been observed in a number of papers that the market return is not alwaysstatistically significant and/or important as a risk factor when other economic variables are consid-ered simultaneously. In this paper, macroeconomic variables are introduced as risk factors along withthe market return, with special emphasis on exchange risk factors since asset returns from a smallopen economy, Canada, are considered. The two approaches mentioned above are used. The Fama andMacbeth (1973) cross-sectional analysis is first performed to identify factors that could be of impor-tance as well as sub-periods where different macro variables might have played a significant role. Thesimultaneous equation analysis is then performed with the chosen economic factors and sub-periods.

The paper proceeds as follows. In the next section, the cross-sectional model is introduced and theestimation results are presented. Section 3 characterizes and reports results from the simultaneousmodel. Finally, conclusions are drawn in the last section.

2. Evaluating risk premia: a cross-sectional model

In this section, a cross-sectional model, similar to the one proposed in Fama and Macbeth (1973), ispresented. The estimation procedure is divided in two steps. In the first stage of the analysis, the betasof the model, the multiple regression coefficients related to a set of chosen variables, are generatedusing a rolling regressions method. The length of the rolling window is fixed. In the second stage, theestimated betas themselves are used as regressors in a cross-section of returns of varying sizes. Thecorresponding risk premia are then estimated.

The first step can be described by the following set of time series regressions:

Rit =∑

jbjiZjt−1 + �it i = 1, . . . , K (1)

where Rit is the period t return on portfolio i in excess of the risk free rate, the Zjt−1’s are theJ information variables used to generate the betas, and �it is the part of each excess return that isorthogonal to the proposed economic variables. There are K portfolio excess returns. A rolling windowof pre-determined length N is used to generate a vector of betas for each time period. The informational

L. Samson / Research in International Business and Finance 28 (2013) 35– 44 37

content of the economic variables is therefore renewed slowly, one observation at a time, allowingthe estimated betas to vary over time.

In the second step, the estimated ˇji’s from Eq. (1), are used in a cross-section of returns, allowingthe price of each factor to be assessed. A factor that is significantly priced by the data has a corre-sponding statistically significant risk premium. Therefore for each period �, the following cross-sectionregression is performed:

Rt = a0t +∑

jajt bjt + �t (2)

where R� is a vector of size portfolio excess returns for period �. The estimated aj� is the riskpremium associated with the corresponding period-� beta. The error term is the part of each excessreturn that is uncorrelated with the constant and the identified risk factors. If a risk factor is priced,the associated estimated aj� ’s will be statistically different from zero.

The analysis is performed using monthly data from the Toronto Stock Exchange (TSE), from 1961:1to 2004:12, a period of forty-four years. The data is regrouped into ten size portfolios. The individualreturns used to form the ten size portfolios are obtained from the Canadian Financial Markets ResearchCenter Database (CFMRC). When calculating the portfolio returns, the weight for each return is keptconstant for the year at its June value. The returns are measured in excess of the 3-month Treasurebill rate, representing the risk free rate.

The economic variables, the Z’s, used to generate the betas and estimate the risk premia are com-parable to those presented in the existing literature.1 The return on the market portfolio (rM), asmeasured by the value weighted return on the TSE, is considered since it is identified by the CAPMliterature as a potentially important risk factor. Previous studies have found inflation and real inter-est rates to play a significant role as well. The first variable will be priced if it can have real effectsvia differing sensitivities or adjustment costs. The second variable summarizes the general state ofinvestment opportunities. The measured inflation rate (�) is the percentage change in the consumerprice index. The real interest rate (r) is the difference between the three-month Treasury bill rate andinflation.

Since forty percent of the goods produced in Canada are sold abroad, and of those, eighty percentare directed towards the United States, a measure of foreign exchange rate risk is also introduced in thelist of relevant economic variables. This variable is defined as the absolute value of the changes in thebilateral Canada–US dollar exchange rate (F). A number of previous studies have found a moderate tosignificant impact of exchange risk on firms’ value for various countries, assuming either asymmetricor symmetric exposure. See for example, Apergis et al. (2011), Doidge et al. (2006), Dominguez andTesar (2006), Goldberg and Veitch (2010), Griffin and Stulz (2001), Koutmos and Martin (2003), Kolariet al. (2008), Virk (2012), Zhao (2010) and others.2

Summary statistics for these variables are presented in Table 1. As expected the smallest size port-folio exhibits the highest average return, but with the corresponding highest standard deviation. Thelargest size portfolio is characterized by the opposite statistics. Among the economic and financial riskfactors, the market return is the most volatile, followed by the exchange rate variable. The real interestrate and the inflation rate are both relatively small and they have comparable standard deviations.The correlations (positive or negative) between these four variables are very small except for the realinterest rate and the inflation rate, with a calculated value of −0.72.3

The first pass of the estimation procedure requires that the length of the rolling window be specifiedwhen estimating Eq. (1). A ten year period, or 120 observations, was chosen. That choice was dictatedby the fact that the period has to be long enough to generate meaningful betas, while not loosing

1 These variables were all obtained from the CANSIM database provided by Statistics Canada.2 Some studies have, on the other hand, found no or very little evidence of a priced exchange risk factor. Examples are Jorion

(1991), Bartov and Bodnar (1994), Griffin and Stulz (2001) and Anatolyev (2008).3 No monthly consumption data is available for Canada, so this variable could not be included, even if the C-CAPM model

would suggest it as a relevant risk factor. Chen et al. (1986) also introduced a production index variable in their list of economicfactors. This series is available monthly, but not on a continuous basis for the entire period considered in this paper.


Table 1Descriptive statistics: returns and economic variables (1971:1–2004:12).

Variable Mean Standard deviation

r1 2.905 9.173r2 1.656 6.553r3 1.150 6.065r4 0.915 5.297r5 0.864 5.147r6 0.865 4.885r7 0.716 4.935r8 0.872 4.862r9 0.899 4.809r10 0.850 4.267rM 0.984 4.867r 0.183 0.415F 0.875 0.720� 0.405 0.421

r1 represents the monthly return for the smallest size portfolio and r10 is the return for the largest firms. The market return (rM)is the value-weighted return on the TSE, (r) is the return on the 3-month Treasury bill less the inflation rate, (F) is the absolutevalue of the changes in the Canada-US dollars exchange rate, and (�) is the CPI inflation rate.

too many observations at the beginning of the sample period. Consequently, the starting date for theestimated betas is January 1971. Four betas per month are estimated this way, one for each risk factor.

The second step of the procedure implies regressing a cross-section of returns on the betas foreach time period, as shown in Eq. (2). Since only ten size portfolios were created (K = 10), due tothe thinness of the Canadian market for small firms, it implies too few degrees of freedom. For thatreason, a pooling of three consecutive months was done when estimating the risk premia, creating adependent variable 30 observations long instead of 10 observations only. This procedure imposes thatthe estimated premia, the aj� ’s, are constant over a three month period for each factor. Instead of 408estimated values for each risk premium, this pooling of data implies that a third of that number, 136,is generated for the 1971:1–2004:12 period.

A multi-factor analysis is performed since it has been shown in previous studies that a variablemight show up as important when considered alone, but may be subsumed by other risk factors in amultivariate regression context. Table 2 reports summary statistics for the estimated premia. The toprow of the table refers to the entire period while the next two rows report statistics related to twosub-samples.

From this table, it appears that exchange risk has been present for the whole sample. The size ofthe estimated premium is stable over time, and in spite of the bias in the standard errors, it seems

Table 2Estimated mean risk premia: cross-sectional model (1971:1–2004:12).

Period Constant rM r F �

1971:1–2004:12 −0.454 0.789 0.209 1.759 0.921(4.059) (30.06) (3.148) (3.542) (4.060)[−1.305] [0.306] [0.775] [5.791] [2.644]

1971:1–1989:12 −0.425 −0.495 0.271 1.716 1.305(3.426) (18.49) (3.294) (3.458) (5.210)[−1.081] [−0.233] [0.718] [4.327] [2.184]

1990:1–2004:12 −0.491 2.414 0.131 1.813 0.433(4.771) (40.35) (2.979) (3.674) (1.667)[−0.798] [0.464] [0.340] [3.822] [2.014]

Estimates of risk premia are obtained from Eq. (2) for a cross-section of the ten size portfolios (maintaining the premia constantfor three months each time), making 136 regressions. Means are calculated over the reported sample periods. Standard errorsin parentheses and Fama and Macbeth t-statistics in brackets.


1971 197 4 197 7 1980 1983 198 6 198 9 199 2 199 5 199 8 200 1 2004-20

-15

-10

-5

0

5

10

15

20

1971 197 4 197 7 1980 1983 198 6 198 9 199 2 199 5 199 8 200 1 2004-10

-5

0

5

10

15

20

A

B

Fig. 1. (A) Estimated exchange risk premium, (B) estimated inflation risk premium.

to play a significant role. Inflation risk has been present mostly in the seventies and eighties, whichcorresponds to the period where central monetary authorities in many economies, including Canada,had to deal with fairly high and volatile inflation rates. By the end of the eighties, inflation was undercontrol in Canada, dictating our choice of sub-sample. The market return premium seems to havebecome predominant in size in the nineties and the following years, but it is estimated with a lot ofimprecision, causing it to be statistically insignificant. The support for the CAPM model is thereforevery weak. The real interest rate does not show up as an important contributing risk factor with anestimated risk premium close to zero for the entire sample period.

Since only the inflation and exchange premia have relatively high t-statistics for the whole sampleperiod, only these two variables are reproduced in Fig. 1A and B. These two premia correspond to theestimated aj� ’s in Eq. (2) associated with the inflation and exchange rate betas. As should be expected,they are both quite volatile. The volatility and mean value of the inflation premium have howeverdecreased substantially after the eighties, while exchange rate risk does not seem to have changedmuch over time.


3. Evaluating risk premia: a simultaneous equations model

A simultaneous equations model with one pre-specified risk factor is presented and tested below.Since the previous analysis has identified a potential role for exchange risk during the entire sampleperiod, this variable Ft is treated as the relevant observed risk factor. The proposed model is similar tothe one presented in Ferson (1990). In the case of one observed risk factor, it is:

Ft − Zt−1 ̨ = �t (3)

R1,t − Zt−1ı − (Ft − Zt−1˛)˝1 = �1,t (4)

Ri,t − (Zt−1ı)(˝1−1˝i) − (Ft − Zt−1˛)˝i − �i,t i = 2, . . . , K (5)

In this system of equations, Ft is the foreign exchange risk factor, �t is the factor realization whichis orthogonal to the information set Zt−1, and the ˝’s are the factor loadings of this realization inthe excess return equations.4 Eq. (3) defines the conditional mean of the relevant state variable, Ft. Assuggested by asset pricing models, it is the innovations in this state variable, the �t’s, that represent therelevant risk for investors. Eq. (3) is necessary to identify these innovations. They represent unexpectedfluctuations in Ft, those that are orthogonal to the information set Zt−1. In this system, portfolio of sizeone is the reference asset and the other size portfolios are the test assets. The excess return on thereference portfolio is characterized by Eq. (4). This regression is a projection of the dependent variableon the innovations in Ft and on the information set. The rest of the system, embodied in Eq. (5), relatesthe mean of the reference asset to those of the test assets through the cross-equation restrictionspresent in the specified model.

The well-known Generalized Method of Moments (GMM) procedure is used to carry out the esti-mations. This method implies defining the following vector of orthogonality conditions:

G(Xt,Zt−1,b) = H(Xt,b) × Zt−1 (6)

Where Xt is the data, Zt−1 is defined above, b is the vector of parameter estimates and H(Xt, b)represents Eqs. (3)–(5). With a well specified model, at the true parameter vector b0, it must be thecase that:

E[G(Xt,Zt−1, b0)] = 0 (7)

The GMM procedure chooses the estimated coefficients so as to be as close as possible to satis-fying condition (7). The reported J-statistics in Table 3 and Table 4 are tests of the over-identifyingrestrictions implied by the model. As indicated by condition (7), they should be close to zero. They aredistributed as Chi-squared statistics.5

Table 3 presents estimation results with the exchange rate variable as the only risk factor. Thelagged exchange rate variable is included in the information set, Zt−1, along with the other previouslymentioned variables, namely the real interest rate, the market return and the inflation rate, to ensurethat the realization of the risk factor is orthogonal to any relevant past information. This table indi-cates that, with a reported J-statistic(46) = 78.19 and a corresponding significance level of 0.002, wellbelow the usual five percent level, the model’s overidentifying restrictions are clearly rejected by thedata when the entire sample period is considered. The factor loadings, the ˝’s, take plausible valueshowever, and are statistically significant. They indicate that small firms (portfolios 2, 3, etc.) may bemore sensitive to exchange risk than larger ones (portfolios . . ., 9, 10).

The assumption of constant loadings for the entire sample period could be too restrictive and maybe causing the model’s rejection by the data. The estimations were therefore carried out for the twosub-samples discussed previously, 1971:1–1989:12 and 1990:1–2004:12. The significance levels aregreatly improved by this split in the sample even if the model’s restrictions are still rejected. The

4 If there are more than one explicit risk factors considered, Ft is replaced by an M-vector of risk factors, and there are K–Mtest assets remaining.

5 For more efficiency, the unpredicted component of exchange rate fluctuations from equation (3) is used as an additionalinstrument. The degrees of freedom of the reported J-Statistics are adjusted accordingly.


Table 3Simultaneous equations model: exchange risk factor, Ft − Zt−1 ̨ = �t , R1,t − Zt−1ı − (Ft − Zt−1˛)˝1 = �1,t ,Ri,t − (Zt−1ı)(˝1

−1˝i) − (Ft − Zt−1˛)˝i − �i,t i = 2, . . ., K.

Period: 1971:1–2004:12

˛0 ˛1 ˛2 ˛3 ˛4 ı0 ı1 ı2 ı3 ı4

0.802 0.003 −0.205 0.257 −0.287 4.326 0.362 −3.083 0.144 −4.615(0.082) (0.006) (0.101) (0.046) (0.117) (0.797) (0.074) (1.016) (0.428) (1.047)

˝2 ˝3 ˝4 ˝5 ˝6 ˝7 ˝8 ˝9 ˝10

0.599 0.439 0.310 0.245 0.188 0.178 0.163 0.129 0.100(0.066) (0.062) (0.054) (0.051) (0.052) (0.055) (0.055) (0.055) (0.052)J-statistic(46): 78.19, significance level: 0.002

Period: 1971:1–1989:12

˛0 ˛1 ˛2 ˛3 ˛4 ı0 ı1 ı2 ı3 ı4

0.368 0.005 0.181 0.201 0.291 4.808 0.275 −5.979 0.222 −4.974(0.087) (0.006) (0.098) (0.055) (0.122) (1.081) (0.070) (1.301) (0.574) (1.357)

˝2 ˝3 ˝4 ˝5 ˝6 ˝7 ˝8 ˝9 ˝10


Period: 1990:1–2004:12

˛0 ˛1 ˛2 ˛3 ˛4 ı0 ı1 ı2 ı3 ı4

1.154 0.005 −0.767 0.218 −0.912 2.850 0.601 0.054 0.027 −1.492(0.116) (0.009) (0.172) (0.058) (0.213) (0.952) (0.120) (1.401) (0.470) (1.350)

˝2 ˝3 ˝4 ˝5 ˝6 ˝7 ˝8 ˝9 ˝10

0.542 0.422 0.240 0.137 0.090 0.090 0.088 0.024 −0.034(0.083) (0.073) (0.060) (0.056) (0.050) (0.057) (0.055) (0.056) (0.052)J-statistic(46): 63.44, significance level: 0.045

Standard errors in parentheses. Variables in Zt−1 are a constant and the lagged values of the market return, the real interest rate,the exchange rate variable and the inflation rate. The ˝’s are the sensitivity of each portfolio’s return to the factor’s realization(relative to portfolio one since, for identification purposes, ˝1 = 1).

estimated loadings are generally larger and slightly more precisely estimated in the first period. Thereported significance level is closer to the usual 5% level for the most recent sub-period however,leading almost to non-rejection of the proposed model. Since the standard errors are not biased whenusing this simultaneous procedure, the previous finding that exchange rate fluctuations have been apriced risk factor appears to be robust. Rejection of the model’s restrictions implies however that thissource of risk is not sufficient to entirely determine the behavior of the analyzed excess returns duringthe period.

The cross-sectional analysis presented in Section II indicated that inflation has potentially playedan important role in the seventies and eighties, a period during which it was relatively high andvolatile. The same analysis revealed that market exposure might, on the other hand, have been a moreimportant contributing risk factor, even if quite imprecise, in the latter part of the sample period.Consequently, we checked for the presence of a statistically significant second risk factor in thosetwo sub-samples. Table 4 reports results of a two-factor model with inflation risk added to exchangerate risk for the 1971:1–1989:12 period, and the market return added as a second risk factor for1990:1–2004:12.6 The exchange rate fluctuations variable is present for both sub-periods since Table 3has shown that, even if it cannot be the sole contributing risk factor, it has nevertheless significantly

6 Estimations were also carried out for the entire period, 1971:1–2004:12, and the alternative sub-samples, for both factors,but did not converge.


Table 4Simultaneous equations model: two risk factors and two sub-periods.

Exchange and inflation risk factors: 1971:1–1989:12

˛0,F ˛1,F ˛2,F ˛3,F ˛4,F ˛0,� ˛1,� ˛2,� ˛3,� ˛4,�

0.416 0.012 0.285 0.087 0.301 0.297 0.001 0.163 0.018 0.433(0.084) (0.006) (0.097) (0.058) (0.118) (0.066) (0.004) (0.080) (0.032) (0.084)

ı0,1 ı1,1 ı2,1 ı3,1 ı4,1 ı0,2 ı1,2 ı2,2 ı3,2 ı4,2

−0.822 0.054 −1.424 0.017 0.621 −0.946 0.073 −1.518 0.141 0.702(0.203) (0.023) (0.668) (0.306) (0.490) (0.242) (0.022) (0.629) (0.329) (0.496)

˝2,F ˝3,F ˝4,F ˝5,F ˝6,F ˝7,F ˝8,F ˝9,F ˝10,F

0.922 0.937 0.332 0.502 0.381 0.368 0.315 0.273 0.774(0.250) (0.244) (0.229) (0.225) (0.214) (0.231) (0.232) (0.246) (0.246)

˝2,� ˝3,� ˝4,� ˝5,� ˝6,� ˝7,� ˝8,� ˝9,� ˝10,�

1.156 1.133 1.062 0.590 0.860 0.541 0.181 0.005 −0.274(0.238) (0.258) (0.264) (0.215) (0.240) (0.219) (0.246) (0.255) (0.344)J-statistic(42): 39.06, significance level: 0.601

Exchange and market risk factors: 1990:1–2004:12

˛0,F ˛1,F ˛2,F ˛3,F ˛4,F ˛0,M ˛1,M ˛2,M ˛3,M ˛4,M

1.059 −0.015 −0.633 0.215 −0.570 −0.944 0.108 0.116 0.349 2.599(0.118) (0.010) (0.176) (0.060) (0.209) (0.655) (0.071) (1.070) (0.266) (1.096)ı0,1 ı1,1 ı2,1 ı3,1 ı4,1 ı0,2 ı1,2 ı2,2 ı3,2 ı4,2

0.691 0.983 3.083 1.048 2.091 −0.739 0.510 1.219 0.706 1.793(0.130) (0.164) (1.120) (0.456) (1.098) (0.375) (0.083) (0.895) (0.275) (0.829)˝2,F ˝3,F ˝4,F ˝5,F ˝6,F ˝7,F ˝8,F ˝9,F ˝10,F

0.216 0.076 −0.081 −0.173 −0.214 −0.242 −0.221 −0.251 −0.314(0.118) (0.117) (0.125) (0.139) (0.132) (0.144) (0.141) (0.125) (0.122)˝2,M ˝3,M ˝4,M ˝5,M ˝6,M ˝7,M ˝8,M ˝9,M ˝10,M


Standard errors in parentheses. Variables in Zt−1 are a constant and the lagged values of the market return, the real interest rate,the exchange rate variable and the inflation rate. The ˝’s are the sensitivity of each portfolio’s return to the factor’s realization(relative to portfolio one since, for identification purposes, ˝1,E = 1, ˝1,� = 1 and ˝1,M = 1).

impacted on excess returns. Portfolios one and two are the reference assets (with unconstrained ıparameters), and the other K-2 portfolios are the test assets.

As shown in the top part of Table 4, adding inflation clearly improves the fit of the model for the firstsub-period. The model’s restrictions are not rejected by the data and both factors are contributing tothis good fit. The factor loadings, the ˝’s, take plausible values and are generally statistically significant.It appears from these estimation results that foreign exchange and inflation risks were both pricedduring the seventies and eighties, especially when small firms are considered.7 Once again, theseresults are generally in support of those presented in Table 2 with the cross-section methodology.

The results for the second sub-period, with the market return as the additional risk factor, tell adifferent story. The model’s restrictions are rejected by the data. The fit is worst than with exchangerate fluctuations as the only risk factor. The loadings, however, indicate that the sensitivity to marketrisk may have been quite important during that period, while the role played by exchange risk isestimated with less precision. But, given the reported significance level of 0.015, these coefficientsmust be interpreted with care. Using the return on the Standard and Poor’s 500 as an alternativedefinition of market return, one implying perfect North American financial integration, did not change

7 A notable exception is portfolio of size 10 composed of the largest Canadian firms, since the exchange rate factor loading(�F) increases for this excess return. This may reflect the fact that these firms represent a large percentage of across-the-bordertransactions.


the basic results.8 No convergence or flat rejection of the model’s restrictions was obtained for thetwo sub-samples (1971:1–1989:12 and 1990:1–2004:12) and the whole period.

4. Conclusion

This paper has investigated the sources of risk present when pricing assets listed on the TorontoStock Exchange, a market from a small open economy. Two methodologies commonly exploited inthe literature were used, a Fama–Macbeth-type cross sectional model and a Ferson-type simultaneousequations with pre-specified factors model. The period under study extends from 1971:1 to 2004:12.The cross-sectional analysis revealed that exchange risk has affected the value of Canadian firms sincethe beginning of the floating exchange rate period in 1970. The important role played by this risk factorhas stayed fairly stable over time. This result suggests that when looking at the costs and benefits ofa floating exchange rate regime, compared to a fixed exchange rate regime or a monetary union, thisimpact should be taken into consideration. The cross-section analysis also revealed that inflation riskhas also played a role in determining the price of assets, mostly in the seventies and eighties, a periodof relatively high and volatile inflation in Canada.

The simultaneous equations model confirmed the important role played by exchange risk indetermining asset returns in Canada. When combined with inflation risk, these two factors led tonon-rejection of the models’ restrictions for the seventies and eighties sub-period. This result canbe interpreted as meaning that, during this period, unexpected inflation and unexpected changesin the bilateral exchange rate were the two most important macroeconomic variables priced by theCanadian market. After the eighties, the exchange risk factor alone implied a nearly non-rejection theover-identifying restrictions of the model. Adding a market return variable as a second risk factor didnot improve the fit of the model implying that very little support was found for the CAPM. In summary,from the analysis performed in this paper, it can be concluded that exchange risk is the only factor,among those considered, that has been consistently priced by the stock market since the beginning ofthe flexible exchange rate regime in Canada.

References

Anatolyev, S., 2008. A 10-year retrospective on the determinants of Russian stock returns. Research in International Businessand Finance 22, 56–67.

Apergis, N., Artikis, P., Sorros, J., 2011. Asset pricing and foreign exchange risk. Research in International Business and Finance25, 308–328.

Bartov, E., Bodnar, G.M., 1994. Firm valuation earnings, expectations, and the exchange-rate exposure effect. Journal of Finance49, 1755–1785.

Chen, N., Roll, R., Ross, S., 1986. Economic forces and the stock market. Journal of Business 59, 383–403.Doidge, C., Griffin, J., Williamson, R., 2006. Measuring the economic importance of exchange rate exposure. Journal of Empirical

Finance 13, 550–576.Dominguez, K.M.E., Tesar, L.L., 2006. Exchange rate exposure. Journal of International Economics 68, 188–218.Elton, E.J., Gruber, M.J., Blake, C.R., 1995. Fundamental economic variables expected returns, and bond fund performance. Journal

of Finance 50, 1229–1256.Fama, E.F., French, K.R., 1992. The cross-section of expected stock returns. Journal of Finance 47, 427–465.Fama, E.F., Macbeth, J., 1973. Risk return and equilibrium: empirical tests. Journal of Political Economy 81, 607–636.Ferson, W.E., 1990. Are the latent variables in time-varying expected returns compensation for consumption risk? Journal of

Finance 45, 397–429.Ferson, W.E., Harvey, C.R., 1991. The variations of economic risk premiums. Journal of Political Economy 99, 385–415.Goldberg, C.S., Veitch, J.M., 2010. Country risk and financial integration: a case study of South Africa. Research in International

Business and Finance 24, 138–145.Griffin, J.M., Stulz, R.M., 2001. International competition and exchange rate shocks: a cross-country industry analysis of stock

returns. Review of Financial Studies 14, 215–241.Harvey, C.R., Solnik, B., Zhou, G., 2002. What determines expected international asset returns. Annals of Economics and Finance

3, 249–298.Jorion, P., 1991. The pricing of exchange rate risk in the stock market. Journal of Financial and Quantitative Analysis 26, 363–

376.

8 For space consideration, these results are not reported here. Using a weighted TSE-SP500 aggregate return would not changethe results since the SP500 would represent approximately 95% of the return.


Kolari, J.W., Moorman, T.C., Sorescu, S.M., 2008. Foreign exchange risk and the cross-section of stock returns. Journal of Inter-national Money and Finance 27, 1074–1097.

Koutmos, G., Martin, A.D., 2003. Asymmetric exchange rate exposure: theory and evidence. Journal of International Money andFinance 22, 365–383.

Shanken, J., Weinsten, M.I., 2006. Economic forces and the stock market revisited. Journal of Empirical Finance 13, 129–144.Virk, N.S., 2012. Stock returns and macro risks: evidence from Finland. Research in International Business and Finance 26, 47–66.Zhao, H., 2010. Dynamic relationship between exchange rate and stock price: evidence from China. Research in International

Business and Finance 24, 103–112.

Documents

Asset prices and exchange risk: Empirical evidence from Canada